def run(self):
self.check_validity_input()
self.set_capture_object()
self.logger.info("running the program")
with open(self._filename, 'w') as file:
# Select boxes
bboxes = []
self.logger.info(HOW_TO)
# OpenCV's selectROI function doesn't work for selecting multiple objects in Python
# So we will call this function in a loop till we are done selecting all objects
while True:
# draw bounding boxes over objects
# selectROI's default behaviour is to draw box starting from the center
# when fromCenter is set to false, you can draw box starting from top left corner
bbox = cv2.selectROI('MultiTracker', self._frame)
bboxes.append(bbox)
k = cv2.waitKey(0) & 0xFF
if k == 113: # q is pressed
if bboxes.__len__() > 2:
break
else:
self.logger.error("\n--At least three objects have to be marked to continue--\n")
cv2.destroyAllWindows()
if bboxes.__len__() % 2 == 0:
self.logger.error("\n--Illegal input. pleas try again.")
sys.exit(1)
self.logger.info('Selected bounding boxes {}'.format(bboxes))
# Create MultiTracker object
multiTracker = cv2.MultiTracker_create()
# Initialize MultiTracker
for bbox in bboxes:
multiTracker.add(cv2.TrackerCSRT_create(), self._frame, bbox)
# Write the number of bbox the user choose
file.write(str(bboxes.__len__()) + "\n")
firstFrame = True
tic = time.perf_counter()
# Process video and track objects
while self._cap.isOpened():
success, frame = self._cap.read()
if not success:
break
# get updated location of objects in subsequent frames
success, boxes = multiTracker.update(frame)
# draw tracked objects
for i, newbox in enumerate(boxes):
center = ((int(newbox[0] + (int(newbox[2]) / 2))) - SHIFTING.X_SHIFTING,
(int(newbox[1] + (int(newbox[3]) / 2))) - SHIFTING.Y_SHIFTING)
file.write(str(center) + "\n")
# some information on processing single frame
if firstFrame:
firstFrame = False
toc = time.perf_counter()
totalFrames = int(self._cap.get(cv2.CAP_PROP_FRAME_COUNT))
self.logger.info("single frame took %.4f seconds" % (toc - tic))
self.logger.info(f"{totalFrames} total frames in video")
self.logger.info("Estimated time to finish: {}".format(
time.strftime('%H:%M:%S', time.gmtime((toc - tic) * totalFrames))))
file.close()
toc = time.perf_counter()
self.logger.info("Finish detection in {}".format(time.strftime('%H:%M:%S', time.gmtime((toc - tic)))))
tryModel.tryModel(self._filename, self._outputPath)
toc = time.perf_counter()
self.logger.info("Finish model in {}".format(time.strftime('%H:%M:%S', time.gmtime((toc - tic)))))
def grab_proc(url, rate, camera_id):
'''
抓图处理进程
:param url:
:param rate:
:param camera_id:
:param logger:
:return:
'''
logger = Log('grab-proc' + str(os.getpid()), 'logs/')
logger.info('初始化seaweedfs')
master = WeedClient(config.get('weed', 'host'),
config.getint('weed', 'port'))
logger.info('初始化Kafka')
kafka = Kafka(bootstrap_servers=config.get('kafka', 'boot_servers'))
topic = config.get('camera', 'topic')
face_tool = Face(config.get('api', 'face_server'))
detect_count = 0 # 用于detect频次计数
frame_internal = track_internal * rate
trackable = False
# 启动抓图线程
q = queue.Queue(maxsize=100)
t = GrabJob(q, camera_id, url, rate,
Log('grab-proc' + str(os.getpid()) + '-thread', 'logs/'),
config)
t.start()
while True:
try:
img = q.get(timeout=20)
if detect_count % frame_internal == 0:
detect_count = 0
b64 = mat_to_base64(img)
t1 = time.time()
detect_result = face_tool.detect(b64)
logger.info('detect cost time: ',
round((time.time() - t1) * 1000), 'ms')
if detect_result['error_message'] != '601':
logger.warning('verifier detector error, error_message:',
detect_result['error_message'])
continue
tracker = cv2.MultiTracker_create()
latest_imgs = []
timestamp = round(time.time())
for face_num in range(detect_result['detect_nums']):
tmp = detect_result['detect'][face_num]
bbox = (tmp['left'], tmp['top'], tmp['width'],
tmp['height'])
tracker.add(cv2.TrackerKCF_create(), img, bbox)
face_b64 = face_tool.crop(bbox[0], bbox[1], bbox[2],
bbox[3], b64, True)
latest_img = {
'image_base64': face_b64,
'bbox': bbox,
'landmark':
detect_result['detect'][face_num]['landmark'],
'time': timestamp
}
# 增加人脸质量过滤
if tmp['sideFace'] == 0 and tmp[
'quality'] == 1 and tmp['score'] > 0.95:
latest_imgs.append(latest_img)
if len(latest_imgs) > 0:
trackable = True
else:
trackable = False
elif trackable:
# 开始追踪
ok, bboxs = tracker.update(img)
if ok and detect_count < frame_internal - 1:
if detect_count % 10 == 0:
logger.info('tracking..., detect_count = %d' %
detect_count)
detect_count += 1
continue
else:
# 取detect到的人脸
logger.info('tracking over! detect_count = %d' %
detect_count)
for latest in latest_imgs:
logger.info([camera_id], 'track person success!')
face_b64 = latest['image_base64']
# save img to seaweed fs
logger.info([camera_id],
'save grabbed detect_result to seaweed fs')
assign = master.assign()
logger.info([camera_id], 'assign result:', assign)
ret = master.upload(assign['url'], assign['fid'],
base64_to_bytes(face_b64),
assign['fid'] + '.jpg')
logger.info([camera_id], 'upload result:', ret)
# send to Kafka
url = 'http' + ':' + '//' + assign[
'url'] + '/' + assign['fid']
logger.info('[', camera_id, ']', 'img url:', url)
msg = json.dumps({
'url': url,
'time': latest['time'],
'camera_id': camera_id,
'landmark': latest['landmark']
})
logger.info([camera_id], 'send to kafka: ', msg)
kafka.send(topic, msg)
# 再次进入detect
detect_count = 0
trackable = False
logger.info('restart detection')
else:
if detect_count % 10 == 0:
logger.info('detect 0 detect_result, do not track',
'detect count= ', detect_count)
detect_count += 1
continue
except queue.Empty:
logger.error('grab queue empty error, exit')
break
detect_count += 1
logger.info('抓图进程终止')
def track(self):
self.countimage = 0
cap = cv2.VideoCapture(self.videopath)
cap.set(1, self.firstFrame)
success, frame = cap.read()
self.video_length = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
if not success:
self.logger.error('Failed to read video')
sys.exit(1)
self.logger.debug(
f'{len(self.bboxes)} tracking boxes have been created')
trackerType = "CSRT"
multiTracker = cv2.MultiTracker_create() # Create MultiTracker object
for bbox in self.bboxes: # Initialize MultiTracker
multiTracker.add(cv2.TrackerCSRT_create(), frame, bbox)
self.logger.info(
f'Starting the tracking of {len(self.bboxes)} particles')
while cap.isOpened():
success, frame = cap.read()
if not success or self.countimage == self.lastFrame:
# cv2.imwrite('img_path{}_lastframe.jpg'.format(self.countimage), previous_frame)
# self.logger.info("\n Last Image save in your working directory !")
break
previous_frame = frame
success, boxes = multiTracker.update(
frame) # get updated location of objects in subsequent frames
if self.countimage < 1: # initialize dataframes type
for i, newbox in enumerate(boxes):
self.positions[i] = []
# draw tracked objects
for i, newbox in enumerate(boxes):
# coordinates of the tracking box
p1 = ((newbox[0]), (newbox[1]))
p2 = ((newbox[0] + newbox[2]), (newbox[1] + newbox[3]))
Pr1 = (int(newbox[0]), int(newbox[1]))
Pr2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(frame, Pr1, Pr2, (255, 0, 0), 2, 1)
# append the center of the box to dict "self.positions[index of box]" according to refresh rate
if self.countimage:
xcenter = (p1[0] + p2[0]) / 2
ycenter = (p1[1] + p2[1]) / 2
self.positions[i].append((xcenter, ycenter))
for pos in self.positions[i]:
xcenter = pos[0]
ycenter = pos[1]
cv2.circle(frame, (int(xcenter), int(ycenter)), 1,
self.circle_color, -1)
self.countimage += 1
cv2.putText(frame, f"Frame no{self.countimage}", (10, 37),
cv2.FONT_HERSHEY_PLAIN, 1, (0, 0, 0), 1)
self.current_frame = frame
yield frame
# quit on ESC button and save last frame
if cv2.waitKey(1) & 0xFF == 27: # Esc pressed
break
def d_t_with_cv_ta(self, stop_thread, format='bgr'):
"""The low-level method to provide detecting and tracking objects with using OpenCV's tracking API.
Args:
stop_thread: Stop flag of the tread about terminating it outside of the function's loop.
format: Color space format.
"""
if self.record:
self.recorder.start("track")
tracked_boxes = [] # this became array. because of overriding.
names = []
multi_tracker = cv2.MultiTracker_create()
rgb, detected_boxes = self.detect_initiate()
found_count = 0
d_t_failure_count = 0
use_detection = 0
for frame in self.camera.capture_continuous(self.raw_capture,
format=format,
use_video_port=True):
if len(detected_boxes) > len(tracked_boxes):
if not self.no_recognize:
names = self.recognize_things(rgb, detected_boxes)
else:
names = None
self.current_frame = cv2.cvtColor(rgb, cv2.COLOR_RGB2BGR)
# Create MultiTracker object
multi_tracker = cv2.MultiTracker_create()
# Initialize MultiTracker
for box in detected_boxes:
# box[3] is x,
# box[0] is y,
# box[1] is x + w,
# box[2] is y + h.
reworked_box = box[3], box[
0], box[1] - box[3], box[2] - box[0]
multi_tracker.add(self.create_tracker_by_name(),
self.current_frame, reworked_box)
found_count += 1
# grab the raw NumPy array representing the image, then initialize the timestamp and occupied/unoccupied text
frame = frame.array
self.current_frame = frame.copy(
) # For reaching with overwrite privileges.
if use_detection >= 3:
rgb, detected_boxes = self.detect_things(self.current_frame)
use_detection = 0
use_detection += 1
# Start timer
timer = cv2.getTickCount()
# get updated location of objects in subsequent frames
is_tracking_success, tracked_boxes = multi_tracker.update(
self.current_frame)
if not len(detected_boxes) >= len(tracked_boxes):
d_t_failure_count += 1
else:
d_t_failure_count = 0
# Calculate Frames per second (FPS)
fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer)
if is_tracking_success and d_t_failure_count < 5:
self.track(self.current_frame, tracked_boxes, names)
elif not is_tracking_success or d_t_failure_count >= 5:
# Tracking failure
cv2.putText(self.current_frame, "Tracking failure detected",
(100, 80), cv2.FONT_HERSHEY_SIMPLEX, 0.75,
(0, 0, 255), 2)
tracked_boxes = [] # for clearing tracked_boxes list.
# # Display tracker type on frame
# cv2.putText(frame, self.tracker_type + " Tracker", (100, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
#
# # Display FPS on frame
# cv2.putText(frame, "FPS : " + str(int(fps)), (100, 50), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
self.show_frame(self.current_frame)
self.truncate_stream()
if self.check_loop_ended(stop_thread):
break
def main(paths,
day=2,
camera=3,
initial_frame=0,
num_frames=36000,
N_pruning=0):
final_frame = initial_frame + num_frames - 1
if num_frames < 1:
logging.info('Number of frames needs to be higher than 0\n')
sys.exit()
if final_frame > 35999:
if initial_frame > 35999:
logging.info('Invalid initial frame (0-35999)\n')
sys.exit()
else:
final_frame = 35999
num_frames = final_frame + 1 - initial_frame
data_path = paths['data_path']
videos_path = paths['videos_path']
output_path = paths['output_path']
if not os.path.isdir(output_path):
logging.info('Output directory does not exist\n')
sys.exit()
# Video to load
video_file = '30min_day' + str(day) + '_cam' + str(
camera) + '_20fps_960x540.MP4'
video_name = 'Day ' + str(day) + ' Camera ' + str(camera)
cap = cv2.VideoCapture(videos_path +
video_file) # Capture object to read video
if not cap.isOpened(): # Exit if video not opened
logging.info('Could not open the video\n')
sys.exit()
cap.set(cv2.CAP_PROP_POS_FRAMES,
initial_frame) # Set the first frame to read
ret, frame = cap.read() # Read first frame
if not ret: # Exit if reading failure
logging.info('Unable to read the video file\n')
sys.exit()
logging.info(f'Selected video: {video_name}\n')
# Read annotations
logging.info('Getting annotations ...\n')
det_full, cam_full, lost_full = read.read_annotations(
path=data_path,
day=day,
initial_frame=initial_frame,
num_frames=num_frames)
# Read annotations from first frame
frame_index = initial_frame
annotations0, num_part0 = read_detections(det=det_full,
cam=cam_full,
lost=lost_full,
camera=camera,
frame_index=frame_index -
initial_frame)
logging.info(
f'{num_part0} participants annotated on the frame {frame_index}\n')
# Object MultiTracker (Primary Trackers)
multi_tracker = cv2.MultiTracker_create()
boxes, targets_tracked = change_det_boxes(
init_boxes=annotations0) # From (x1,y1,x2,y2) to (x1,y1,width,height)
for box in boxes: # Initialization of primary trackers for each target
multi_tracker.add(cv2.TrackerKCF_create(), frame, box)
multi_tracker.add(cv2.TrackerMedianFlow_create(), frame, box)
multi_tracker.add(cv2.TrackerMIL_create(), frame, box)
# MHT
tracking_params = {
'N_pruning': N_pruning, # Index for pruning
'distance_threshold':
100, # Distance threshold for hypothesis formation
'distance_threshold2':
75, # Distance threshold for the updating of primary trackers
'MIL_weight': 0.2, # MIL Tracker weight on the track scoring
'MF_weight': 0.35, # MF Tracker weight on the track scoring
'KCF_weight': 0.45, # KCF Tracker weight on the track scoring
'color_score_threshold':
0.20, # Bhattacharyya distance threshold score between color histograms for Re-ID
'color_score_weight':
0.75, # Color histograms weight on the lost tracks scoring
'lost_time_threshold': 25, # Time of loss threshold for Re-ID
'lost_time_weight':
0.25, # Time of loss weight on the lost tracks scoring
'color_hist_bins': 4
} # Number of bins per histogram
mht = MHT(tracking_params) # Object MHT initialized
logging.info('Running MHT ...\n')
ti = time.time() # Start timer
logging.info(f'Frame: {frame_index} ...')
mht.init(frame=frame, detections=annotations0) # Run MHT on first frame.
ids = [] # ID's of targets tracked at each frame
fps_acc = 0 # Processing speed
trk_path = output_path + 'tracker/'
if not os.path.isdir(trk_path):
os.mkdir(trk_path)
res_file = trk_path + 'Results_day' + str(day) + '_cam' + str(
camera) + '_' + str(tracking_params['N_pruning'])
speed_file = trk_path + 'Speed_day' + str(day) + '_cam' + str(
camera) + '_' + str(tracking_params['N_pruning'])
runtime_file = trk_path + 'Time_day' + str(day) + '_cam' + str(
camera) + '_' + str(tracking_params['N_pruning'])
frame_index += 1
#########################################
frame_print = set(np.arange(initial_frame - 1, final_frame,
100)) # To print frame every 100 frames
frame_save = set(np.arange(initial_frame - 1, final_frame,
1000)) # To save results every 1000 frames
# Process video and track objects
while cap.isOpened() and (frame_index <= final_frame) and (frame_index <=
35999):
ret, frame = cap.read() # Read new frame
if not ret: # Exit if reading failure
break
if frame_index in frame_print:
logging.info(f'Frame: {frame_index} ...')
timer = cv2.getTickCount() # Start timer to get FPS
# Read annotations
annotations, num_part = read_detections(det=det_full,
cam=cam_full,
lost=lost_full,
camera=camera,
frame_index=frame_index -
initial_frame)
if num_part < num_part0:
logging.info(
f'Number of annotations changed from {num_part0} to {num_part} on frame {frame_index}\n'
)
num_part0 = num_part
if num_part > num_part0:
logging.info(
f'Number of annotations changed from {num_part0} to {num_part} on frame {frame_index}\n'
)
num_part0 = num_part
# Update primary trackers for the current frame
ret, multitracker_results = multi_tracker.update(frame)
# Turn results from primary trackers into a dictionary for MHT
trackers_results = change_track_boxes(init_boxes=multitracker_results,
indexes=targets_tracked)
# Run MHT with annotations (detections) and tracker results
solution_coord, track_ids, new_tracks = mht.run(
frame=frame,
detections=annotations,
trackers_results=trackers_results)
# Update primary trackers when they're lost or there are new targets
if len(new_tracks) != 0:
for key in new_tracks.keys():
if key in ids: # If an existing tracker needs to be updated
i = targets_tracked.index(key)
targets_tracked[i] = random.randint(50, 500) # Change ID
targets_tracked.append(key) # Append corresponding ID
# Append new trackers for the target
box = new_tracks[key]
new_box = (box[0], box[1], box[2] - box[0], box[3] - box[1])
multi_tracker.add(cv2.TrackerKCF_create(), frame, new_box)
multi_tracker.add(cv2.TrackerMedianFlow_create(), frame,
new_box)
multi_tracker.add(cv2.TrackerMIL_create(), frame, new_box)
ids = track_ids # Update track ID's
fps = cv2.getTickFrequency() / (
cv2.getTickCount() - timer
) # Compute frames per second (FPS) of the processing
fps_acc += fps
if frame_index in frame_save: # Save results
tf = time.time() # End timer
t_tot = tf - ti
fps_mean = fps_acc / (frame_index - initial_frame + 1
) # Average FPS
logging.info(f'Elapsed time: {tf-ti} seconds\n')
logging.info(
f'{frame_index-initial_frame+1} frames processed ({initial_frame}-{frame_index})\n'
)
# Save results to a .CSV file
#results_file = res_file+'_'+str(initial_frame)+'-'+str(frame_index)+'.csv'
#write_csv(file_name=results_file, solution_coordinates=solution_coord)
fps_file = speed_file + '_' + str(initial_frame) + '-' + str(
frame_index) + '.csv'
np.savetxt(fps_file, [fps_mean], delimiter=',')
time_file = runtime_file + '_' + str(initial_frame) + '-' + str(
frame_index) + '.csv'
np.savetxt(time_file, [t_tot], delimiter=',')
frame_index += 1
tf = time.time() # End timer
t_tot = tf - ti
fps_mean = fps_acc / (frame_index - initial_frame) # Average FPS
logging.info('MHT completed\n')
logging.info(f'Elapsed time: {t_tot} seconds\n')
logging.info(
f'{frame_index-initial_frame} frames processed ({initial_frame}-{frame_index-1})\n'
)
# Save results to a .CSV file
results_file = res_file + '_' + str(initial_frame) + '-' + str(
frame_index - 1) + '.csv'
write_csv(file_name=results_file, solution_coordinates=solution_coord)
fps_file = speed_file + '_' + str(initial_frame) + '-' + str(frame_index -
1) + '.csv'
np.savetxt(fps_file, [fps_mean], delimiter=',')
time_file = runtime_file + '_' + str(initial_frame) + '-' + str(
frame_index - 1) + '.csv'
np.savetxt(time_file, [t_tot], delimiter=',')
def detect(input_video, output_video, interval, track_number, csv_path,
mode_path, debuglog):
"""
input_video:输入视频的路径
output_video:输出结果视频的路径
interval:检测是隔多少帧检测一次
csv_path:输出csv文件的路径
"""
# 提示，输入三个要检测的目标
debuglog('Select {} tracking targets'.format(track_number))
cv2.namedWindow("tracking")
# 输入要检测视频的名称和路径
camera = cv2.VideoCapture(input_video)
tracker = cv2.MultiTracker_create()
init_once = False
model = load_model(mode_path)
ok, image = camera.read()
# 设置保存视频结果的信息
fourcc = cv2.VideoWriter_fourcc(*'MJPG')
out = cv2.VideoWriter(output_video, fourcc, 20,
(image.shape[1], image.shape[0]))
if not ok:
debuglog('Failed to read video')
exit()
# 设置跟踪目标的个数
track_number = track_number
# 把目标跟踪的框放入空列表
bboxes = []
for t in range(track_number):
bbox = cv2.selectROI('tracking', image) #第一个框用来检测时间参数
bboxes.append(bbox)
i = 1
#保存气孔面积的列表
aliens = []
while camera.isOpened():
ok, image = camera.read()
if not ok:
debuglog('no image to read')
camera.release()
cv2.destroyAllWindows()
break
# 每隔指定的间隔帧数才进行语义分割
if i % interval == 0:
if not init_once:
for bbox in bboxes:
ok = tracker.add(cv2.TrackerBoosting_create(), image, bbox)
init_once = True
ok, boxes = tracker.update(image)
# print(ok, boxes)
j = 0
# 将一帧里的气孔面积保存在一个列表中
stoma_area = []
for newbox in boxes:
#判断，如果是第一个框，则检测时间参数
if j == 0:
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]),
int(newbox[1] + newbox[3]))
cut_img = cut(image, newbox)
time_text = recognition_number(cut_img)
# 如果不是第一帧，则进行语义分割
else:
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]),
int(newbox[1] + newbox[3]))
cut_img = cut(image, newbox)
# print(cut_img.shape)
seg_img = segment_img(cut_img, model)
# print(seg_img.shape)
# 将检测结果与原始图片混合相加
dst = cv2.addWeighted(cut_img, 0.5, seg_img, 0.5, 0)
image[int(newbox[1]):int(newbox[1] + newbox[3]),
# 将混加的结果放到原图上去
int(newbox[0]):int(newbox[0] + newbox[2])] = dst
# 画矩形
cv2.rectangle(image, p1, p2, (200, 0, 0))
# 显示数字
# 将掩膜变成灰度图像
gray_img = cv2.cvtColor(seg_img, cv2.COLOR_BGR2GRAY)
area_number = calculate_area(gray_img)
stoma_area.append(area_number)
text = '{}'.format(area_number)
cv2.putText(image, text,
(int(newbox[0]), int(newbox[1] - 5)),
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255),
2)
j = j + 1
# stoma_area变成字典
dict_keys = []
dict_values = stoma_area
for s in range(len(dict_values)):
# 生成keys的列表
dict_keys.append('stomat{}'.format(s))
# 生成气孔的字典
new_dict1 = dict(zip(dict_keys, dict_values))
# 生成序号和时间的字典
new_dict2 = {'image_number': i, 'time_date': time_text}
# 合并字典
new_aliens = dict(new_dict1, **new_dict2)
aliens.append(new_aliens)
cv2.imshow('tracking', image)
out.write(image)
debuglog('这是第{}张图片！'.format(i))
else:
pass
k = cv2.waitKey(1)
if k == 27 & 0xFF == ord('q'):
break #按q退出
i = i + 1
out.release()
# 将字典转化为pandas
df = pd.DataFrame(aliens)
#保存为excel表格
df.to_csv(csv_path)
_, frame = cap.read() # Capturamos o primeiro frame
bb = [] # Criamos uma lista vazia que receberá as coordenadas dos boxes
out = cv2.VideoWriter('teste_out.mp4', fourcc, 30.0, (frame.shape[1], frame.shape[0])) # Determina o nome do arquivo de saída, sua taxa de FPS e sua resolução.
while True:
roi = cv2.selectROI('Frame', frame) # Função para seleção de ROI
#print(roi)
bb.append(roi) # Coordenadas do box da ROI adicionadas à lista
k = cv2.waitKey(0)
if k == ord('q'):
break
multiTracker = cv2.MultiTracker_create() # Cria o objeto Tracker
for bbox in bb:
multiTracker.add(get_tracker(), frame, bbox) # Inicializa o objeto Tracker para cada ROI selecionada
while True:
old_frame = frame
ret, frame = cap.read() # Captura um frame
if not ret: # Verifica status do vídeo
exit()
_, bxs = multiTracker.update(frame) # Atualiza o objeto Tracker para a nova posição de cada ROI selecionada
for ID, box in enumerate(bxs):
def mot_detect(args):
def nothing(emp):
pass
print("----------> create the trackers")
# extract the OpenCV version info
(major, minor) = cv2.__version__.split(".")[:2]
# if we are using OpenCV 3.2 OR BEFORE, we can use a special factory
# function to create our object tracker
if int(major) == 3 and int(minor) < 3:
tracker = cv2.Tracker_create(args["tracker"].upper())
# otherwise, for OpenCV 3.3 OR NEWER, we need to explicity call the
# approrpiate object tracker constructor:
else:
# initialize a dictionary that maps strings to their corresponding
# OpenCV object tracker implementations
OPENCV_OBJECT_TRACKERS = {
"csrt": cv2.TrackerCSRT_create,
"kcf": cv2.TrackerKCF_create,
"boosting": cv2.TrackerBoosting_create,
"mil": cv2.TrackerMIL_create,
"tld": cv2.TrackerTLD_create,
"medianflow": cv2.TrackerMedianFlow_create,
"mosse": cv2.TrackerMOSSE_create
}
# grab the appropriate object tracker using our dictionary of
# OpenCV object tracker objects
trackers = cv2.MultiTracker_create()
# tracker = OPENCV_OBJECT_TRACKERS[args["tracker"]]()
# initialize the bounding box coordinates of the object we are going
# to track
initBB = None
tclass = []
start_time = time.time()
failure_flag = False
# if a video path was not supplied, grab the reference to the web cam
if not args.get("video", False):
print("[INFO] starting video stream...")
vs = VideoStream(src=0).start()
time.sleep(1.0)
# otherwise, grab a reference to the video file
else:
vs = cv2.VideoCapture(args["video"])
# initialize the FPS throughput estimator
fps = None
frames = int(vs.get(cv2.CAP_PROP_FRAME_COUNT))
loop_flag = 0
pos = 0
print("----------> read the frames")
cv2.namedWindow('Frame')
cv2.createTrackbar('time', 'Frame', 0, frames, nothing)
# loop over frames from the video stream
while True:
# process-bar setting
if loop_flag == pos:
loop_flag = loop_flag + 1
cv2.setTrackbarPos('time', 'Frame', loop_flag)
else:
pos = cv2.getTrackbarPos('time', 'Frame')
loop_flag = pos
vs.set(cv2.CAP_PROP_POS_FRAMES, pos)
# grab the current frame, then handle if we are using a
# VideoStream or VideoCapture object
frame = vs.read()
frame = frame[1] if args.get("video", False) else frame
# check to see if we have reached the end of the stream
if frame is None:
break
# resize the frame (so we can process it faster) and grab the
# frame dimensions
frame = imutils.resize(frame,
width=int(args["set_width"]),
height=int(args["set_height"]))
frame0 = frame.copy()
(H, W) = frame.shape[:2]
# check to see if we are currently tracking an object
if initBB is not None:
print("----------> update the trackers' roi area")
# grab the new bounding box coordinates of the object
(success, boxes) = trackers.update(frame)
print("[INFO] success / box num", success, len(boxes))
# check to see if the tracking was a success
if success:
tpos = []
for num, box in enumerate(boxes):
(x, y, w, h) = [int(v) for v in box]
new_pos = [x, y, x + w, y + h]
tpos.append(new_pos)
cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 0),
2)
cv2.putText(frame, tclass[num], (x, y - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2)
now_time = time.time()
if now_time - start_time > float(
args["interval"]) and len(boxes) > 0:
start_time = now_time
pic_name = str(time.time()) + ".jpg"
pic_fullpath = Path(args["pic_dir"]).joinpath(pic_name)
print("[INFO] save new pic:", pic_fullpath)
r = Image.fromarray(frame0[:, :, 2]).convert('L')
g = Image.fromarray(frame0[:, :, 1]).convert('L')
b = Image.fromarray(frame0[:, :, 0]).convert('L')
img = Image.merge("RGB", (r, g, b))
print("----------> save the pic and annotation xml")
img.save(pic_fullpath)
annotation_single_img(args["pic_dir"], pic_name,
args["xml_dir"], tclass, tpos)
# update the FPS counter
fps.update()
fps.stop()
# initialize the set of information we'll be displaying on
# the frame
info = [
("Tracker", args["tracker"]),
("Success", "Yes" if success else "No"),
("FPS", "{:.2f}".format(fps.fps())),
]
# loop over the info tuples and draw them on our frame
for (i, (k, v)) in enumerate(info):
text = "{}: {}".format(k, v)
cv2.putText(frame, text, (10, H - ((i * 20) + 20)),
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2)
else:
trackers.clear()
trackers = cv2.MultiTracker_create()
failure_flag = True
initBB = None
tclass = []
cv2.putText(frame, lost_warning, (10, 100),
cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2)
# show the output frame
cv2.imshow("Frame", frame)
if failure_flag:
cv2.waitKey(5000) & 0xFF
failure_flag = False
else:
key = cv2.waitKey(100) & 0xFF
# if the 's' key is selected, we are going to "select" a bounding
# box to track
if key == ord("s"):
print("----------> select roi by mouse")
cv2.putText(frame, input_cmd, (10, 20), cv2.FONT_HERSHEY_SIMPLEX,
0.6, (0, 0, 255), 2)
cv2.imshow("Frame", frame)
# show the classes choice
ckey = cv2.waitKey(0) & 0xFF
if int(ckey - 48) > len(select_classes) or int(ckey - 48) <= 0:
continue
cname = select_classes[int(ckey - 48)]
print("[INFO] choose type to label:", cname)
# select the bounding box of the object we want to track (make
# sure you press ENTER or SPACE after selecting the ROI)
initBB = cv2.selectROIs("Frame",
frame,
fromCenter=False,
showCrosshair=True)
# start OpenCV object tracker using the supplied bounding box
# coordinates, then start the FPS throughput estimator as well
initBB = tuple(map(tuple, initBB))
if str(initBB) == '()':
print("[WARNING] There is no select ROIs!")
# initBB==None
else:
for bb in initBB:
tracker = OPENCV_OBJECT_TRACKERS[args["tracker"]]()
trackers.add(tracker, frame, bb)
tclass.append(cname)
fps = FPS().start()
# if the `q` key was pressed, break from the loop
elif key == ord("q"):
print("----------> quit all process")
break
elif key == ord("r"):
print("----------> clear all roi trackers")
tclass = []
trackers.clear()
trackers = cv2.MultiTracker_create()
cv2.putText(frame, input_cmd, (10, 20), cv2.FONT_HERSHEY_SIMPLEX,
0.6, (0, 0, 255), 2)
cv2.imshow("Frame", frame)
# show the classes choice
ckey = cv2.waitKey(0) & 0xFF
cname = select_classes[int(ckey - 48)]
print("[INFO]You have chosen the class:%s" % cname)
initBB = cv2.selectROIs("Frame",
frame,
fromCenter=False,
showCrosshair=True)
initBB = tuple(map(tuple, initBB))
if str(initBB) == '()':
print("[WARNING] There is no select ROIs!")
# initBB==None
else:
print("---------->add new roi trackers")
for bb in initBB:
tracker = OPENCV_OBJECT_TRACKERS[args["tracker"]]()
trackers.add(tracker, frame, bb)
tclass.append(cname)
# else:
# continue
# if we are using a webcam, release the pointer
if not args.get("video", False):
vs.stop()
# otherwise, release the file pointer
else:
vs.release()
# close all windows
cv2.destroyAllWindows()
def gen_keypoint(self):
params = self.set_params()
multiTracker = cv2.MultiTracker_create()
bboxes = []
colors = [(255, 0, 0), (0, 255, 0)]
try:
# Starting OpenPose
opWrapper = self.op.WrapperPython()
opWrapper.configure(params)
opWrapper.start()
stream = cv2.VideoCapture(self.input_video)
#stream.set(cv2.CAP_PROP_FPS, 2)
frame_width = int(stream.get(3))
frame_height = int(stream.get(4))
out = cv2.VideoWriter(self.output_vdieo,
cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'),
10, (frame_width, frame_height))
# Process Video
poseKeypoints_data_video = []
map_bone_video = []
while True:
datum = self.op.Datum()
ret, img = stream.read()
if ret == False:
break
datum.cvInputData = img
opWrapper.emplaceAndPop([datum])
# Display Image
#print("Body keypoints: \n" + str(datum.poseKeypoints) + str(type(datum.poseKeypoints)))
poseKeypoints_data = []
show_result = datum.cvOutputData
if len(datum.poseKeypoints.shape) > 1:
for poseKeypoints in datum.poseKeypoints:
bone_struct = dict()
for i in range(len(poseKeypoints)):
keypoint = dict()
keypoint['x'] = poseKeypoints[i][0]
keypoint['y'] = poseKeypoints[i][1]
#keypoint['c'] = str(poseKeypoints[i][2])
bone_struct[i] = keypoint
#self.resize_bone(bone_struct,self.config.fixed_bone_size)
poseKeypoints_data.append(bone_struct)
print("cal_size: ", cal_size_bone(bone_struct))
#draw_bone(bone_struct,cv2,show_result)
if len(bboxes) == 0:
for bst in poseKeypoints_data:
bbox = box_from_bonestruct(bst)
bboxes.append(bbox)
colors.append((randint(0, 255), randint(0, 255),
randint(0, 255)))
for bbox in bboxes:
multiTracker.add(cv2.TrackerCSRT_create(), img,
bbox)
poseKeypoints_data_video.append(poseKeypoints_data)
print(len(poseKeypoints_data_video))
# if len(poseKeypoints_data_video) == 5:
# print("IMPM")
# multiTracker.add(cv2.TrackerCSRT_create(), img, (100, 300, 500, 700))
# colors.append((randint(0, 255), randint(0, 255), randint(0, 255)))
success, boxes = multiTracker.update(img)
map_bone = map_bone_struct_to_bbox(poseKeypoints_data, boxes)
map_bone_video.append(map_bone)
for i in range(len(poseKeypoints_data)):
if map_bone[i] == -1 and check_bone_struct(
poseKeypoints_data[i]):
bbox = box_from_bonestruct(poseKeypoints_data[i])
bboxes.append(bbox)
color = (randint(0,
255), randint(0,
255), randint(0, 255))
colors.append(color)
multiTracker.add(cv2.TrackerCSRT_create(), img, bbox)
map_bone[i] = len(colors) - 1
draw_bone_with_color(poseKeypoints_data[i], cv2,
show_result, colors[int(map_bone[i])])
for i, newbox in enumerate(boxes):
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]),
int(newbox[1] + newbox[3]))
print(p1, p2)
cv2.rectangle(show_result, p1, p2, colors[i], 2, 1)
cv2.imshow("OpenPose OUPUT VIDEO", show_result)
out.write(show_result)
print("YYYYYYYYYYYYYYYYYYYYYYYYYYEEEEE")
key = cv2.waitKey(1)
if key == ord('q'):
break
print("XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX")
stream.release()
out.release()
cv2.destroyAllWindows()
return [poseKeypoints_data_video, map_bone_video]
except Exception as e:
print(e)
def process(args):
objects_detected = dict()
#ToDo: Put this in intermediate_detection
tracker_types = ['BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN']
tracker_type = tracker_types[2]
tracker = None
"""
if tracker_type == 'BOOSTING':
tracker = cv.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
tracker = cv.TrackerGOTURN_create()
"""
predictor = object_detector(args.model, args.config)
multi_tracker = cv.MultiTracker_create()
stream = cv.VideoCapture(args.input if args.input else 0)
window_name = "Tracking in progress"
cv.namedWindow(window_name, cv.WINDOW_NORMAL)
cv.setWindowProperty(window_name, cv.WND_PROP_AUTOSIZE, cv.WINDOW_AUTOSIZE)
cv.moveWindow(window_name, 10, 10)
if args.output:
_, test_frame = stream.read()
height = test_frame.shape[0]
width = test_frame.shape[1]
fourcc = cv.VideoWriter_fourcc(*'XVID')
#out = cv.VideoWriter(args.output,fourcc, 20.0, (640,480))
out = cv.VideoWriter(args.output, fourcc, 20.0, (width, height))
failTolerance = 0
if args.classes:
with open(args.classes, 'rt') as f:
classes = f.read().rstrip('\n').split('\n')
else:
classes = list(np.arange(0, 100))
stream, objects_detected, objects_list, multi_tracker = intermediate_detections(
stream, predictor, multi_tracker, tracker, args.thr, classes)
while stream.isOpened():
grabbed, frame = stream.read()
if not grabbed:
break
timer = cv.getTickCount()
#Even when multitracker fails, bboxes will have old values
#But ok will be false
if len(objects_list) > 0:
ok, bboxes = multi_tracker.update(frame)
#bboxes = multi_tracker.getObjects()
#ok = multi_tracker.empty()
fps = cv.getTickFrequency() / (cv.getTickCount() - timer)
print(bboxes, ' --- ', ok)
#if ok and len(bboxes) > 0:
#if ok and len(bboxes) > 0:
if ok and len(bboxes) > 0:
drawPred(frame, bboxes, objects_detected)
# Display FPS on frame
cv.putText(frame, "FPS : " + str(int(fps)), (100, 50),
cv.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
else:
cv.putText(frame,
'Tracking Failure. Trying to detect more objects',
(50, 80), cv.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
stream, objects_detected, objects_list, multi_tracker = intermediate_detections(
stream, predictor, multi_tracker, tracker, args.thr, classes)
# Display result
#If resolution is too big, resize the video
if frame.shape[1] > 1240:
cv.imshow(window_name, cv.resize(frame, (1240, 960)))
else:
cv.imshow(window_name, frame)
#Write to output file
if args.output:
out.write(frame)
k = cv.waitKey(1) & 0xff
#Force detect new objects if 'q' is pressed
if k == ord('q'):
print('Refreshing. Detecting New objects')
cv.putText(frame, 'Refreshing. Detecting New objects', (100, 80),
cv.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
stream, objects_detected, objects_list, multi_tracker = intermediate_detections(
stream, predictor, multi_tracker, tracker, args.thr, classes)
# Exit if ESC pressed
if k == 27: break
stream.release()
if args.output:
out.release()
cv.destroyAllWindows()
def __d_t_with_cv_ta(self, stop_thread, format='bgr'):
"""Method to provide detecting and tracking objects with using OpenCV's tracking API.
Args:
stop_thread: Stop flag of the tread about terminating it outside of the function's loop.
format: Color space format.
"""
tracked_boxes = [] # this became array. because of overriding.
names = []
multi_tracker = cv2.MultiTracker_create()
rgb, detected_boxes = self.__detect_initiate()
found_count = 0
d_t_failure_count = 0
use_detection = 0
last_frame = np.zeros(shape=(self.frame_height, self.frame_width))
while True:
if last_frame.any() != self.current_frame.any():
last_frame = self.current_frame
if len(detected_boxes) > len(tracked_boxes):
if not self.no_recognize:
names = self.__recognize_things(rgb, detected_boxes)
else:
names = None
last_frame = cv2.cvtColor(rgb, cv2.COLOR_RGB2BGR)
# Create MultiTracker object
multi_tracker = cv2.MultiTracker_create()
# Initialize MultiTracker
for box in detected_boxes:
# box[3] is x,
# box[0] is y,
# box[1] is x + w,
# box[2] is y + h.
reworked_box = box[3], box[0], box[1] - box[3], box[2] - box[0]
multi_tracker.add(self.__create_tracker_by_name(), last_frame, reworked_box)
found_count += 1
if use_detection >= 3:
rgb, detected_boxes = self.detect_things(last_frame)
use_detection = 0
use_detection += 1
# Start timer
timer = cv2.getTickCount()
# get updated location of objects in subsequent frames
is_tracking_success, tracked_boxes = multi_tracker.update(last_frame)
if not len(detected_boxes) >= len(tracked_boxes):
d_t_failure_count += 1
else:
d_t_failure_count = 0
# Calculate Frames per second (FPS)
fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer)
if is_tracking_success and d_t_failure_count < 5:
self.track(last_frame, tracked_boxes, names)
elif not is_tracking_success or d_t_failure_count >= 5:
# Tracking failure
cv2.putText(last_frame, "Tracking failure detected", (100, 80), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
tracked_boxes = [] # for clearing tracked_boxes list.
# # Display tracker type on frame
# cv2.putText(frame, self.tracker_type + " Tracker", (100, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
#
# # Display FPS on frame
# cv2.putText(frame, "FPS : " + str(int(fps)), (100, 50), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
self.__show_frame(last_frame)
# self.__truncate_stream()
if self.__check_loop_ended(stop_thread):
break
def multi_tracking(yolo, video_path):
global percentResize
accum_time = 0
curr_fps = 0
fps = "FPS: ??"
prev_time = timer()
global multi_tracking_enable
global arima_predict
multi_tracking_enable = True
tracker = cv2.MultiTracker_create()
camera = cv2.VideoCapture(video_path)
ok, image = camera.read()
if not ok:
print('Failed to read video')
exit()
boxes = []
frameCount = 1
startFrame = 1
skipFrames = 25
consecutiveframes = 1
initialHistoryCount = 11
skipHistory = 5
extrapolate = 3
dof = 0
yoloCount = 0
countCond = 0
xhistory = []
yhistory = []
depth_history = []
while(True):
if(frameCount == startFrame):
frame = Image.fromarray(image)
frame, boxes = yolo.detect_image(frame)
#yolo.close_session()
break
ok, image = camera.read()
frameCount += 1
#np.set_printoptions(suppress = True)
boxes = np.asarray(boxes)
boxes[:, 2] = boxes[:, 2] - boxes[:, 0]
boxes[:, 3] = boxes[:, 3] - boxes[:, 1]
boxes = np.ndarray.tolist(boxes)
prevBoxes = len(boxes)
curBoxes = prevBoxes
#print(boxes)
#np.savetxt('boxes.txt', boxes, fmt = "%i")
#return boxes
#boxes = []
manualBox = 0
for i in range(manualBox):
box = cv2.selectROI('tracking', image)
boxes.append(box)
for eachBox in boxes:
eachBox = tuple(eachBox)
xhistory.append([int(eachBox[0] + eachBox[2] / 2)])
yhistory.append([int(eachBox[1] + eachBox[3] / 2)])
ok = tracker.add(cv2.TrackerMedianFlow_create(), image, eachBox)
while(True):
ok, image=camera.read()
if not ok:
break
orig_image = image.copy()
if(prevBoxes != curBoxes):
countCond += 1
if(frameCount % skipFrames == 0):
#print(consecutiveframes)
consecutiveframes = 1
tracker = cv2.MultiTracker_create()
frame = Image.fromarray(image)
boxes = []
frame, boxes = yolo.detect_image(frame)
yoloCount += 1
boxes = np.asarray(boxes)
boxes[:, 2] = boxes[:, 2] - boxes[:, 0]
boxes[:, 3] = boxes[:, 3] - boxes[:, 1]
boxes = np.ndarray.tolist(boxes)
prevBoxes = len(boxes)
curBoxes = None
xhistory = []
yhistory = []
depth_history = []
for eachBox in boxes:
eachBox = tuple(eachBox)
xhistory.append([int(eachBox[0] + eachBox[2] / 2)])
yhistory.append([int(eachBox[1] + eachBox[3] / 2)])
ok = tracker.add(cv2.TrackerMedianFlow_create(), image, eachBox)
#frameCount += 1
#continue
ok, boxes = tracker.update(image)
for i in range(len(boxes)):
xhistory[i].append(int(boxes[i][0] + boxes[i][2] / 2))
yhistory[i].append(int(boxes[i][1] + boxes[i][3] / 2))
if(arima_predict and len(xhistory[0]) > initialHistoryCount):
#if(len(xhistory[i]) > 27): dof = 5
#print(xhistory[0])
for i in range(len(boxes)):
history = xhistory[i].copy()
history = [xhistory[i][t] for t in range(0, len(xhistory[i]), skipHistory)]
xmin = min(history)
history[:] = [x - xmin for x in history]
xmax = max(history)
if(xmax == 0): xmax = 1
history[:] = [x / xmax for x in history]
#print('xh', len(history))
for j in range(extrapolate):
xmodel = ARIMA(history, order = (dof, 1, 0))
xmodel_fit = xmodel.fit(disp = 0, maxiter=200)
xoutput = xmodel_fit.forecast()
history.append(xoutput[0])
xhat = int((xoutput[0] * xmax) + xmin)
#xhat = xoutput[0]
history = yhistory[i].copy()
history = [yhistory[i][t] for t in range(0, len(yhistory[i]), skipHistory)]
ymin = min(history)
history[:] = [y - ymin for y in history]
#history = [yhistory[i][0], yhistory[i][int(len(yhistory[i]) / 2)], yhistory[i][len(yhistory[i]) - 1]]
ymax= max(history)
if(ymax == 0): ymax = 1
history[:] = [y / ymax for y in history]
#print('yh', len(history))
for j in range(extrapolate):
ymodel = ARIMA(history, order = (dof, 1, 0))
ymodel_fit = ymodel.fit(disp = 0, maxiter=200)
youtput = ymodel_fit.forecast()
history.append(youtput[0])
yhat = int((youtput[0] * ymax) + ymin)
#yhat = youtput[0]
cp1 = int(boxes[i][0] + boxes[i][2] / 2)
cp2 = int(boxes[i][1] + boxes[i][3] / 2)
cv2.arrowedLine(image, (int(xhistory[i][0]),int(yhistory[i][0])), (cp1, cp2), (0, 255, 0), 2)
cv2.arrowedLine(image, (cp1, cp2), (xhat, yhat), (0, 0, 255), 2)
#slope = math.abs(math.atan((yhat - cp2) / (xhat - cp1)))
#speed = math.sqrt((yhat - cp2) * (yhat - cp2) + (xhat - cp1) * (xhat - cp1))
#percentChange = 0.0
#if(yhat >= cp2):
p1 = (int(xhat - boxes[i][2] / 2), int(yhat - boxes[i][3] / 2))
p2 = (int(xhat + boxes[i][2] / 2), int(yhat + boxes[i][3] / 2))
cv2.rectangle(image, p1, p2, (255, 255, 255), 1)
for newbox in boxes:
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(image, p1, p2, (200,0,0), 2)
if(depthMapEstimation):
depth_est = image_depth(orig_image)
dof = 0
current_depth_est = depth_est.copy()
pred_depth_est = depth_est.copy()
pd = 'OFF'
for i in range(len(boxes)):
p1 = (int(boxes[i][0]), int(boxes[i][1]))
p2 = (int(boxes[i][0] + boxes[i][2]), int(boxes[i][1] + boxes[i][3]))
current_depth = cal_depth_box(depth_est, p1, p2)
if(len(depth_history) < len(boxes)):
depth_history.append([current_depth])
else:
depth_history[i].append(current_depth)
if(math.isnan(current_depth)):
continue
if(len(depth_history[i]) > initialHistoryCount):
pd = 'ON'
history = depth_history[i].copy()
hisotry = np.nan_to_num(history)
history = [history[t] for t in range(0, len(history), skipHistory)]
dmin = min(history)
history[:] = [d - dmin for d in history]
dmax = max(history)
if(dmax == 0): dmax = 1
history[:] = [d / dmax for d in history]
for j in range(extrapolate):
dmodel = ARIMA(history, order = (0, 1, 0))
dmodel_fit = dmodel.fit(disp = 0, maxiter=200)
doutput = dmodel_fit.forecast()
history.append(doutput[0])
#print(doutput[0])
if(not math.isnan(doutput[0])):
dhat = int((doutput[0] * dmax) + dmin)
else:
dhat = current_depth
current_depth_est = set_depth(current_depth_est, p1, p2, current_depth)
if(math.isnan(current_depth)):
print("wtf just happened")
cv2.putText(current_depth_est,text=str(int(current_depth)), org=(p1[0], p1[1]), fontFace=cv2.FONT_HERSHEY_SIMPLEX,
fontScale=0.50, color=(0, 0, 255), thickness=1)
pred_depth_est = set_depth(pred_depth_est, p1, p2, dhat)
cv2.putText(pred_depth_est,text=str(int(dhat)), org=(p1[0], p1[1]), fontFace=cv2.FONT_HERSHEY_SIMPLEX,
fontScale=0.50, color=(0, 0, 255), thickness=1)
cv2.putText(pred_depth_est, text=pd, org=(3, 15), fontFace=cv2.FONT_HERSHEY_SIMPLEX,
fontScale=0.50, color=(0, 0, 255), thickness=2)
#cv2.namedWindow("curdepth", cv2.WINDOW_NORMAL)
current_depth_est = cv2.resize(current_depth_est, (0,0), fx=percentResize, fy=percentResize)
cv2.imshow('curdepth', current_depth_est)
#cv2.namedWindow("predepth", cv2.WINDOW_NORMAL)
pred_depth_est = cv2.resize(pred_depth_est, (0,0), fx=percentResize, fy=percentResize)
cv2.imshow('predepth', pred_depth_est)
curr_time = timer()
exec_time = curr_time - prev_time
prev_time = curr_time
accum_time = accum_time + exec_time
curr_fps = curr_fps + 1
if accum_time > 1:
accum_time = accum_time - 1
fps = "FPS: " + str(curr_fps)
curr_fps = 0
cv2.putText(image, text=fps, org=(3, 15), fontFace=cv2.FONT_HERSHEY_SIMPLEX,
fontScale=0.50, color=(255, 0, 0), thickness=2)
#cv2.namedWindow("tracking", cv2.WINDOW_NORMAL)
image = cv2.resize(image, (0,0), fx=percentResize, fy=percentResize)
cv2.imshow('tracking', image)
frameCount += 1
consecutiveframes += 1
k = cv2.waitKey(1)
if k == 27 : break # esc pressed
print(yoloCount)
print(countCond)
yolo.close_session()
def vid_main(vid1, vid2, fps):
# Open video file for read.
cap1 = cv2.VideoCapture(vid1)
cap2 = cv2.VideoCapture(vid2)
# Perpare dataset and recognition models
FRmodel, ml_model, encoder = rs.set_up(align_method, ml_method)
print("Successfully set up recongnition system.")
# Determine homography transfroamtion equal for all scene if cameras are stable.
if SC:
H = homo_stationary_cameras(vid2, vid1, ratio, reprojThresh,
reprojThresh2)
# Initialize.
count = 0
new_scene = False
new_scene_count = 0
ppl_counts = [0, 0, 0, 0] # number of people prediction in each sub-frame
names_0, names_1, names_2, names_3 = (
[], [], [], []) # list of name prediction in each sub-frame
object_trackers = []
all_names = []
names_colors = {
} # each person with a pre-collected dataset and is detected in the frame is marked by an unique colour.
sub_frames = [] # len(sub_frames)=1 if FP==True, else=4
while cap1.isOpened() and cap2.isOpened(
): #take minimum length of the two videos
print(count) # Current frame number
ret1, frame1 = cap1.read()
ret2, frame2 = cap2.read()
if ret1 and ret2:
# At first frame, obtain input frame dimension and create VideoWriter object.
if count == 0:
height1, width1, layers1 = frame1.shape
height2, width2, layers2 = frame2.shape
new_height = int(height1 + height2 * outputSizeTimes)
new_width = int(width1 + width2 * outputSizeTimes)
# Coordinate up-left coordinate so that frame1 with (height1, width1)
# will be placed at the centre of (new_height, new_width).
new_coord = [
int((new_width - width1) / 2.),
int((new_height - height1) / 2.)
]
# Initialize the dimension of output video format.
frame_width, frame_height = display.output_shape(
PB, FP, new_width, new_height, maxWidth, maxHeight)
out = cv2.VideoWriter(output_file_name,
cv2.VideoWriter_fourcc(*'MP4V'), fps,
(frame_width, frame_height))
# Initialize sub-frames.
mask_img = np.zeros((new_height, new_width)).astype(np.uint8)
mask_img[new_coord[1]:new_coord[1] + height1,
new_coord[0]:new_coord[0] + width1].fill(255)
canvas = np.zeros(
(new_height, new_width,
3)).astype(np.uint8) # dimension of sub-frame
if canvas_color != (0, 0, 0):
canvas[:, :, :] = canvas_color
# SCENE DETECTION:
# Check video scene change, assuming the two cameras cut at the same time.
# Else the sorresponding frames do not match.
pre_frame1 = frame1 #count == 0
new_frame1_norm_gray = cv2.cvtColor(
(frame1 * 255.0 / frame1.max()).astype('uint8'),
cv2.COLOR_BGR2GRAY)
pre_frame1_norm_gray = cv2.cvtColor(
(pre_frame1 * 255.0 / pre_frame1.max()).astype('uint8'),
cv2.COLOR_BGR2GRAY)
curr_err = mse_maxpool(new_frame1_norm_gray, pre_frame1_norm_gray)
if count == 0:
pre_err = curr_err
curr_err_diff = abs(pre_err - curr_err)
if count == 0:
pre_err_diff = curr_err_diff
new_scene = identify_scene(pre_err_diff, curr_err_diff, pre_err,
curr_err, cutThreshold,
aroundCutThreshold)
# Update for the next *necognition* process.
if new_scene or count == 0:
new_scene_count = cutTextDuration
object_trackers = []
all_names = []
ppl_counts = [0, 0, 0, 0]
# pre_frame1 = frame1###
# PANORAMA FEATURE DETECTION:
# Use the last calculated frame's homography transformation.
if count % perFrame == 0:
# Create PANORAMA of the two frame.
# Create describer.
(kps1, features1) = detectAndDescribe(frame1)
(kps2, features2) = detectAndDescribe(frame2)
# Machtes.
M = matchKeypoints(kps2, kps1, features2, features1, ratio,
reprojThresh)
# If the match is None, then there aren't enough matched keypoints to create a panorama
# Assume the views are not overlapped, display both videos by simply paste them together as default.
if M is None and not SC:
result_img = display.homo_not_match_subframe(
PR, canvas, new_width, new_height, new_coord, frame1,
frame2)
common_frame = canvas.copy()
common_frame[:, :, :] = (0, 0, 0)
else:
if not SC: # Evaluate individual frame's homography transformation when the cameras are not stationary.
(matches, H, status) = M
# Create panorama.
(panorama, frame2_trans,
pano_coord) = warp_images(frame1, frame2, H)
# Coordinate for transfering panorama coordinates result to a proper sized canvas,
# with frame1 at the centre.
pano_bounds, new_bounds = display.pano_trans_bounds(
pano_coord, new_coord, new_width, new_height, width1,
height1, panorama.shape)
pano_left_bound, pano_right_bound, pano_up_bound, pano_down_bound = pano_bounds
new_left_bound, new_right_bound, new_up_bound, new_down_bound = new_bounds
# frame2 after homogrpahy transformation, for poisson blending and determine the common region.
target_img = canvas.copy()
target_img[new_up_bound:new_down_bound, new_left_bound:new_right_bound, :] =\
frame2_trans[pano_up_bound:pano_down_bound, pano_left_bound:pano_right_bound, :]
# POISSON BLENDING:
if PB:
# Input images for poisson blending.
source_img = canvas.copy()
source_img[new_coord[1]:new_coord[1] + frame1.shape[0],
new_coord[0]:new_coord[0] +
frame1.shape[1], :] = frame1
# Resize the images.
source_img = cv2.resize(source_img,
(reduce_width, reduce_height))
target_img = cv2.resize(target_img,
(reduce_width, reduce_height))
mask_img = cv2.resize(mask_img,
(reduce_width, reduce_height))
# Apply Poisson blending.
result_img = poisson_blending(source_img, target_img,
mask_img)
else:
result_img = canvas.copy()
result_img[new_up_bound:new_down_bound, new_left_bound:new_right_bound, :] =\
panorama[pano_up_bound:pano_down_bound, pano_left_bound:pano_right_bound, :]
# Mask indicate the common area of the two frames.
common_mask = canvas.copy()
common_mask[:, :, :] = (0, 0, 0)
common_mask[target_img == 0] = 255
common_mask[mask_img == 0] = 255
# Apply to output frame.
common_frame = result_img.copy()
common_frame[common_mask == 255] = 0
# Find and label faces in each sub-frame.
if FP:
sub_frames = [panorama, common_frame, frame1, frame2]
# cv2.imwrite("sub1.png", panorama) ## Save INDIVIDUAL FRAMES for test purpose
# cv2.imwrite("sub2.png", common_frame)
# cv2.imwrite("sub3.png", frame1)
# cv2.imwrite("sub4.png", frame2)
else:
sub_frames = [panorama]
for i, sub_img in enumerate(sub_frames):
# Re-perform detection and recognation when a video scene change is detected,
# then identify people by tracking.
if new_scene or count == 0:
### Replaced by recognition: >>>
# # FACE DETECTION:
# faces, eyes = mtcnn_detect(new_frame)
### <<<
# FACE RECOGNITION with DETECTION
names, patches, faces = rec.predict(
sub_img, ml_model, encoder, align_method, FRmodel)
ppl_counts[i] = len(names)
for name in names:
if name not in names_colors:
names_colors[name] = (randint(0, 255),
randint(0, 255),
randint(0, 255))
# OBJECT TRACKING:
# Create MultiTracker for tracking multiple objects.
curr_multiTracker = cv2.MultiTracker_create()
for (x, y, w, h) in faces:
curr_multiTracker.add(createTrackerByName(trackerType),
sub_img, (x, y, w, h))
object_trackers.append(curr_multiTracker)
all_names.append(names)
new_scene = False
else: # not new scene: previous detected scene = current frame scene
# Update face tracking using previouly initialized multiple object tracker.
success, faces = object_trackers[i].update(sub_img)
display.label_faces(sub_img, faces, all_names[i], names_colors)
# Panorama /main sub-frame
if i == 0:
result_img[new_up_bound:new_down_bound, new_left_bound:new_right_bound, :] =\
sub_img[pano_up_bound:pano_down_bound, pano_left_bound:pano_right_bound, :] #update labels
# cv2.imwrite("sub5.png", result_img) ## Save LABELED MAIN FRAME for test purpose
# Define new frame.
new_frame = display.format_output_frame(FP, canvas, new_coord,
result_img, common_frame,
frame1, frame2)
# Add text anotations:
# Display the "New scene" sign in new_scene_count frame, shortly after new cut detection.
if new_scene_count >= 0:
new_frame = display.display_new_scene_text(
new_frame, new_width, new_height)
# Display text for number of people.
new_frame = display.display_ppl_num_text_ul(
new_frame, new_width, ppl_counts[0], ppl_counts[1],
ppl_counts[2], ppl_counts[3]) # Final count: up-left image.
if FP:
new_frame = display.display_ppl_num_text_ur(
new_frame, new_width, ppl_counts[1]) # up-right image.
new_frame = display.display_ppl_num_text_bl(
new_frame, new_width, new_height,
ppl_counts[2]) # bottom-left image.
new_frame = display.display_ppl_num_text_br(
new_frame, new_width, new_height,
ppl_counts[3]) # bottom-right image.
# if count==0:
# cv2.imwrite("test.png", new_frame) ## Save FIRST FRAME of output video for test purpose.
# Write to output video.
out.write(new_frame)
# Update for the next *frame*.
pre_frame1_norm_gray = new_frame1_norm_gray.copy()
pre_err = curr_err
pre_err_diff = curr_err_diff
new_scene_count -= 1
names_0, names_1, names_2, names_3 = ([], [], [], [])
sub_frames = []
count += 1
elif cv2.waitKey(1) & 0xFF == ord('q'): # quit on ESC button
break
else:
break
# Release everything if job is finished
cap1.release()
cap2.release()
print("Video analysis complete!")
out.release()
cv2.destroyAllWindows() # destroy all the opened windows
return
def track(self):
if self.videopath == -1:
self.logger.error('Videopath error')
return("VideopathError")
num_of_img = 0
countimage = 0
cap = cv2.VideoCapture(self.videopath)
success, frame = cap.read()# Read first frame
self.video_length = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
if not success:
self.logger.error('Failed to read video')
sys.exit(1)
bboxes = []
colors = []
a = 2
self.bboxes = []
for circle in self.circles:
x = circle[0]
y = circle[1]
radius = circle[2]
width = radius * 2
height = width
self.bboxes.append( (x-(radius+a),y-(radius+a), width, height))
self.logger.debug(f'{len(self.bboxes)} tracking boxes have been created')
trackerType = "CSRT"
multiTracker = cv2.MultiTracker_create()# Create MultiTracker object
for bbox in self.bboxes: # Initialize MultiTracker
multiTracker.add(cv2.TrackerCSRT_create(), frame, bbox)
self.logger.info(f'Starting the tracking of {len(self.bboxes)} particles')
while cap.isOpened():
success, frame = cap.read()
if not success:
cv2.imwrite('img_path{}_lastframe.jpg'.format(num_of_img), lastframe)
num_of_img += 1
self.looger.info("\n Last Image save in your working directory !")
break
lastframe = frame
success, boxes = multiTracker.update(frame) # get updated location of objects in subsequent frames
if countimage < 1: # initialize dataframes type
for i, newbox in enumerate(boxes):
self.positions[i] = []
# draw tracked objects
for i, newbox in enumerate(boxes):
# coordinates of the tracking box
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(frame, p1, p2, (255, 0, 0), 2, 1)
# append the center of the box to dict "self.positions[index of box]" according to refresh rate
if countimage:
xcenter = (p1[0] + p2[0]) / 2
ycenter = (p1[1] + p2[1]) / 2
self.positions[i].append((xcenter, ycenter))
for pos in self.positions[i]:
xcenter = pos[0]
ycenter = pos[1]
cv2.circle(frame, (int(xcenter), int(ycenter)), 1, self.circle_color, -1)
countimage += 1
cv2.putText(frame, f"Frame no{countimage}", (10, 37), cv2.FONT_HERSHEY_PLAIN, 1, (0, 0, 0), 1)
cv2.imshow('Analysis', frame)
# quit on ESC button and save last frame
if cv2.waitKey(1) & 0xFF == 27: # Esc pressed
cv2.imwrite('img_path{}_lastframe.jpg'.format(num_of_img), frame)
num_of_img += 1
self.logger.info("\n Last Image save in your working directory !")
break
print_progress(countimage, self.video_length)
return(self.positions)
import numpy as np
import cv2 as cv
import sys
# if len(sys.argv) != 2:
# print('Input video name is missing')
# exit()
print('Select 3 tracking targets')
cv.namedWindow("tracking")
camera = cv.VideoCapture(0)
tracker = cv.MultiTracker_create()
init_once = False
ok, image = camera.read()
if not ok:
print('Failed to read video')
exit()
bbox1 = cv.selectROI('tracking', image)
bbox2 = cv.selectROI('tracking', image)
# bbox3 = cv.selectROI('tracking', image)
while camera.isOpened():
ok, image = camera.read()
if not ok:
print('no image to read')
break
if not init_once:
def get_features(self): # type: () -> List[Feature]
_ = self.image.value
rospy.sleep(2)
trackers = {col: cv2.MultiTracker_create() for col in self.masks.keys()}
# Wait for image to warm up
image = self.image.wait_for_n_messages(10)
features = {}
for col, mask in self._split_image_into_masks(image).items():
boxes = self._find_boxes_to_track(mask)
track_image = cv2.cvtColor(mask.astype(np.uint8) * 255, cv2.COLOR_GRAY2BGR)
# print('ADDING BOXES: ', boxes)
for box in boxes:
trackers[col].add(make_tracker('Boosting'), track_image, box)
# print(boxes)
features[col] = [[] for _ in boxes]
duration = 2
rate = rospy.Rate(100)
start_time = rospy.get_time()
while not rospy.is_shutdown() and rospy.get_time() - start_time < duration:
image = self.image.value
show_image = image
for col, mask in self._split_image_into_masks(image).items():
try:
track_image = cv2.cvtColor(mask.astype(np.uint8) * 255, cv2.COLOR_GRAY2BGR)
success, boxes = trackers[col].update(track_image)
boxes = [(box[0] - 10, box[1] - 10, box[2] + 20, box[3] + 20) for box in boxes]
# print(success, boxes)
for box in boxes:
x, y, w, h = (int(v) for v in box)
show_image = cv2.rectangle(show_image, (x, y), (x+w, y+h), (255, 0, 0), 5)
features_found = self._find_features(mask, boxes, col)
for index, feature in enumerate(features_found):
features[col][index].append(feature)
for feature in features_found:
if feature is not None:
show_image = cv2.drawContours(show_image, [feature.contour], -1, (255, 255, 0), 5)
col = self.col_name_to_rgb(feature.colour)
# print(col)
show_image = cv2.putText(
show_image,
'{}'.format(feature.shape),
tuple(int(x) for x in feature.centroid),
cv2.FONT_HERSHEY_SIMPLEX,
3.0,
col,
)
except Exception as err:
print(err)
self.gui_image_pub.publish(self.bridge.cv2_to_imgmsg(show_image))
rate.sleep()
# cv2.waitKey(0)
# print(features)
features = sum(([self._combine_features(f) for f in features[col]] for col in self.masks.keys()), [])
return [f for f in features if f is not None]
def trackMobject(self):
trackerTypes = ['BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN', 'MOSSE', 'CSRT']
def createTrackerByName(trackerType):
# Create a tracker based on tracker name
if trackerType == trackerTypes[0]:
tracker = cv2.TrackerBoosting_create()
elif trackerType == trackerTypes[1]:
tracker = cv2.TrackerMIL_create()
elif trackerType == trackerTypes[2]:
tracker = cv2.TrackerKCF_create()
elif trackerType == trackerTypes[3]:
tracker = cv2.TrackerTLD_create()
elif trackerType == trackerTypes[4]:
tracker = cv2.TrackerMedianFlow_create()
elif trackerType == trackerTypes[5]:
tracker = cv2.TrackerGOTURN_create()
elif trackerType == trackerTypes[6]:
tracker = cv2.TrackerMOSSE_create()
elif trackerType == trackerTypes[7]:
tracker = cv2.TrackerCSRT_create()
else:
tracker = None
print('Incorrect tracker name')
print('Available trackers are:')
for t in trackerTypes:
print(t)
return tracker
if __name__ == '__main__':
print("Default tracking algoritm is CSRT \n"
"Available tracking algorithms are:\n")
for t in trackerTypes:
print(t)
trackerType = "CSRT"
# Set video to load
self.video_track = self.ids['video_track']
videofile = self.video_track.text
#videoPath = 'videofile'
# Create a video capture object to read videos
cap = cv2.VideoCapture(videofile)
# Read first frame
success, frame = cap.read()
# quit if unable to read the video file
if not success:
print('Failed to read video')
sys.exit(1)
## Select boxes
bboxes = []
colors = []
# OpenCV's selectROI function doesn't work for selecting multiple objects in Python
# So we will call this function in a loop till we are done selecting all objects
while True:
# draw bounding boxes over objects
# selectROI's default behaviour is to draw box starting from the center
# when fromCenter is set to false, you can draw box starting from top left corner
bbox = cv2.selectROI('multiTracker/MultiTracker', frame)
bboxes.append(bbox)
colors.append((randint(64, 255), randint(64, 255), randint(64, 255)))
print("Press q to quit selecting boxes and start tracking")
print("Press any other key to select next object")
k = cv2.waitKey(0) & 0xFF
if (k == 113): # q is pressed
break
print('Selected bounding boxes {}'.format(bboxes))
## Initialize MultiTracker
# There are two ways you can initialize multitracker
# 1. tracker = cv2.MultiTracker("CSRT")
# All the trackers added to this multitracker
# will use CSRT algorithm as default
# 2. tracker = cv2.MultiTracker()
# No default algorithm specified
# Initialize MultiTracker with tracking algo
# Specify tracker type
# Create MultiTracker object
multiTracker = cv2.MultiTracker_create()
# Initialize MultiTracker
for bbox in bboxes:
multiTracker.add(createTrackerByName(trackerType), frame, bbox)
# Process video and track objects
while cap.isOpened():
success, frame = cap.read()
if not success:
break
# get updated location of objects in subsequent frames
success, boxes = multiTracker.update(frame)
# draw tracked objects
for i, newbox in enumerate(boxes):
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(frame, p1, p2, colors[i], 2, 1)
# show frame
cv2.imshow('multiTracker/MultiTracker', frame)
# quit on ESC button
k = cv2.waitKey(5) & 0xFF
if k == 27:
break
def main():
camera, net = initializer(VIDEO, PROTO, MODEL)
init_once = False
tracker = None
boxes = tuple()
ok, frame = camera.read()
if not ok:
print('Failed to read video')
exit()
fr_h, fr_w = frame.shape[:2]
cnt = 0
while camera.isOpened():
ok, frame = camera.read()
if not ok:
print('no frame to read')
break
# get prediction on current frame
detections = forward_net(frame, net)
# Display objects count
for j, (obj, obj_cnt) in enumerate(gen_obj_cnt(detections)):
cv2.putText(img=frame,
text='{}: {}'.format(obj, obj_cnt),
org=(5, 40 * (j + 1)),
fontFace=cv2.FONT_HERSHEY_SIMPLEX,
fontScale=1,
color=(0, 0, 255),
thickness=2)
if cnt % 5 == 0:
del tracker
# concatenate a new box. Ref:
# https://stackoverflow.com/a/22732845/8809592
# boxes = np.vstack([boxes, np.array([[770, 590, 90, 80]])]) if len(boxes) else np.array(
# [[580, 570, 100, 90]])
boxes = get_trk_box(detections, fr_h, fr_w)
print(boxes)
for box in boxes:
box[0] = box[0] - cnt * 2
box[1] = box[1] - cnt * 2
tracker = cv2.MultiTracker_create()
init_once = False
if not init_once:
for i, box in enumerate(boxes):
if not (0 < box[0] < fr_w) & (0 < box[1] < fr_h) & (box[0] + box[2] < fr_w) & (box[1] + box[3] < fr_h) \
& (box[2] != 0) & (box[3] != 0):
boxes = np.delete(boxes,
np.where(np.all(boxes == box, axis=1)),
0)
else:
tracker.add(cv2.TrackerMIL_create(), frame, tuple(box))
init_once = True
ok, boxes = tracker.update(frame)
for tkr_box in list(boxes):
p1 = (int(tkr_box[0]), int(tkr_box[1]))
p2 = (int(tkr_box[0] + tkr_box[2]), int(tkr_box[1] + tkr_box[3]))
cv2.rectangle(frame, p1, p2, (200, 0, 0))
cv2.imshow('tracking', frame)
k = cv2.waitKey(1)
if k == 27:
break # esc pressed
cnt += 1
def run(video, pts_src, sport):
Player_Marked = False
# otherwise, grab a reference to the video file
i = 0
FPS_SMOOTHING = 0.9
fps = 0.0
prev = 0
distance = 0
total_distance = 0
speed_list = []
max_speed = 0
min_speed = 0
pointer = 0
t = 3
frames_count = 0
if (sport == 1):
per_frame = 90
t = 3
else:
per_frame = 60
t = 2
# loop over frames from the video stream
vs = cv2.VideoCapture(video)
#change_res(vs,854,480)
trackers = cv2.MultiTracker_create()
while True:
frames_count = frames_count + 1
# grab the current frame, then handle if we are using a
# VideoStream or VideoCapture object
frame = vs.read()
frame = frame[1]
# now = time.time()
# fps = (fps*FPS_SMOOTHING + (1/(now - prev))*(1.0 - FPS_SMOOTHING))
# prev = now
# fpstext = 'FPS = ' + str(int(fps))
##real_time_stats_on_screen
real_time(frame, distance, total_distance, max_speed, min_speed)
# check to see if we have reached the end of the stream
if frame is None:
#print(corr)
average_speed = sum(speed_list) / len(speed_list)
return xs, ys, frames, total_distance, max_speed, min_speed, average_speed
break
# resize the frame (so we can process it faster)
#frame = imutils.resize(frame, width=1920,height=1080)
# grab the updated bounding box coordinates (if any) for each
# object that is being tracked
(success, boxes) = trackers.update(frame)
# loop over the bounding boxes and draw then on the frame
for box in boxes:
(x, y, w, h) = [int(v) for v in box]
cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 255), 2)
# print(str(x+int(w/2))+','+str(y+int(h/2)))
if (i % per_frame == 0):
# xs.append(x+int(w/2))
# ys.append(y+int(h))
x_map, y_map = map.map([x + int(w / 2), y + int(h)], pts_src,
sport)
xs.append(x_map)
ys.append(y_map)
frames.append(frames_count)
points.append(x_map)
points.append(y_map)
if (len(points) >= 4):
#(x1,y1,x2,y2)
#distance = mt.distance(points[pointer-3],points[pointer-2],points[pointer-1],points[pointer])/100
distance = dis.calculateDistance(points[pointer - 3],
points[pointer - 2],
points[pointer - 1],
points[pointer], sport)
total_distance = total_distance + distance
speed = speedFunction.calculate_Speed(distance, t)
speed_list.append(speed)
max_speed = max(speed_list)
min_speed = min(speed_list)
pointer += 2
cv2.circle(frame, (x + int(w / 2), y + int(h)), 5, red, -1)
i += 1
cv2.imshow("Frame", frame)
key = cv2.waitKey(1) & 0xFF
# if the 's' key is selected, we are going to "select" a bounding
# box to track
if (Player_Marked == False):
if key == ord("s"):
# select the bounding box of the object we want to track (make
# sure you press ENTER or SPACE after selecting the ROI)
# mt.mappingMatch()
# mt.mappingPlayer()
box = cv2.selectROI("Frame",
frame,
fromCenter=False,
showCrosshair=True)
# create a new object tracker for the bounding box and add it
# to our multi-object tracker
tracker = OPENCV_OBJECT_TRACKERS['csrt']()
trackers.add(tracker, frame, box)
Player_Marked = True
# if the `q` key was pressed, break from the loop
if key == ord("q"):
break
if key == ord("p"):
cv2.waitKey()
# otherwise, release the file pointer
else:
vs.release()
# close all windows
cv2.destroyAllWindows()
def flow (videoPath, trackerType, labelName, savefolder, waitingTime) :
if __name__ == '__main__' :
print("\n\n\nrun module separately \n\n\n")
from BBoxToDataset.CreateTracker import createTrackerByName
from BBoxToDataset.BboxObject import BboxObject
from BBoxToDataset.BboxObject import FileData
else :
print("\n\n\ncalled by\n", __name__, "\n\n")
from .BBoxToDataset.CreateTracker import createTrackerByName
from .BBoxToDataset.BboxObject import BboxObject
from .BBoxToDataset.BboxObject import FileData
# Create a video capture object to read videos
cap = cv2.VideoCapture(videoPath)
# Read first frame
success, frame = cap.read()
# quit if unable to read the video file
if not success:
print('Failed to read video')
sys.exit(1)
#namedWindow : string 이름을 가진 window 를 표시한다.
#imshow : 특정 window 에 img 를 표시한다.
cv2.namedWindow('Selecting RoI')
frame = cv2.resize(frame, dsize=(416, 416), interpolation=cv2.INTER_AREA)
cv2.imshow('Selecting RoI', frame)
print("size is : ", np.size(frame, 0), np.size(frame, 1))
## Select boxes
bboxes = []
colors = []
# OpenCV's selectROI function doesn't work for selecting multiple objects in Python
# So we will call this function in a loop till we are done selecting all objects
count = 0
while True:
# draw bounding boxes over objects
# selectROI's default behaviour is to draw box starting from the center
# when fromCenter is set to false, you can draw box starting from top left corner
print("\n\n현재 저장된 boundary box 개수 : {}".format(count))
bbox = cv2.selectROI('Selecting RoI', frame)
print("\n\n")
print("object 선택을 마치고 tracking 을 시작하려면 q를 누르세요.")
print("다음 object 를 선택하려면 아무 키나 누르세요.")
bboxes.append(bbox)
colors.append((randint(0, 255), randint(0, 255), randint(0, 255)))
count += 1
k = cv2.waitKey(0) & 0xFF
if (k == 113): # q is pressed
break
print('Selected bounding boxes : {}'.format(bboxes))
# Create MultiTracker object
multiTracker = cv2.MultiTracker_create()
# Initialize MultiTracker
for bbox in bboxes:
multiTracker.add(createTrackerByName(trackerType), frame, bbox)
#----------------------------------------#
#Show original boundingbox with Subwindow#
#----------------------------------------#
frameandBbox = frame
for bbox, color in zip(bboxes, colors):
p1 = (int(bbox[0]), int(bbox[1]))
p2 = (int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3]))
cv2.rectangle(frameandBbox, pt1 = p1, pt2 = p2, color = color, thickness= 2)
print(p1, p2, color)
cv2.namedWindow('Selecting RoI')
cv2.imshow('Selecting RoI', frameandBbox)
#---------------------#
#Movie Saving Settings#
#---------------------#
w = np.size(frame, 0)
h = np.size(frame, 0)
output_size = (int(w), int(h))
print(output_size)
codec = cv2.VideoWriter_fourcc('m', 'p', '4', 'v')
outconfig = cv2.VideoWriter('%s_output.mp4' % (videoPath.split('.')[0]), codec, cap.get(cv2.CAP_PROP_FPS), output_size)
#cap.get(cv2.CAP_PROP_FPS) : get - cap 에서 가져와. prop_fps : 적당한 frame per second를. 즉, 그 동영상의 fps
#output_size : 위에서 설정한 hyper parameter 임.
#-----------------------------------------#
#Imagefile and Annotatioin Saving Settings#
#-----------------------------------------#
# bbox object 객체들을 보관할것
# bbox 가 4개라면 bbox 객체 4개를 만드는것
bboxobjects = []
for i in range(0,len(bboxes),1) :
a = BboxObject((0,0), (0,0), labelName)
bboxobjects.append(a)
if not os.path.isdir(savefolder) :
# save folder 이 존재하지 않는다면..
os.makedirs(savefolder)
print("폴더가 존재하지 않아, 새로운 폴더를 생성합니다.")
# 폴더를 생성한다.
#class init parameters
filefullpath = videoPath
savefolder = savefolder
#-------------------------------#
#Process video and track objects#
#-------------------------------#
font = cv2.FONT_HERSHEY_SIMPLEX #just font setting
save_image_per_n_milliseconds = waitingTime
framecount = 0
while cap.isOpened():
success, frame = cap.read()
if not success:
break
framecount += 1
frame = cv2.resize(frame, dsize=(416, 416), interpolation=cv2.INTER_AREA)
originalframe = frame.copy()
filedata = FileData(filefullpath, framecount, frame, savefolder, labelName)
# get updated location of objects in subsequent frames
success, boxes = multiTracker.update(frame)
# 그냥 박스 안에 무슨 요소들이 들어가 있는지 궁금해서...
# print(boxes)
# draw tracked objects
# 만약 object 가 4개라면 loop 를 4번 돌게 된다.
for i, newbox in enumerate(boxes):
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(frame, p1, p2, colors[i], 2, 1)
# Set Font and put Text on BBox
cv2.putText(frame, labelName,
(int(newbox[0]), int(newbox[1] - 4)),
font,
fontScale=0.3,
color=colors[i],
thickness=1,
lineType=cv2.LINE_AA)
if framecount % save_image_per_n_milliseconds == 0 :
# -ms 로 지정한 만큼마다, bbox object 객체의 데이터를 초기화해줌.
# 초기화된 데이터는 object 객체에 들어가줘야함.
bboxobjects[i].__init__(p1, p2, labelName)
filedata.setObject(bboxobjects[i])
cv2.rectangle(frame,
(p1[0]-7,p1[1]-7),
(p2[0]+7,p2[1]+7),
colors[i], 1, 1)
# show frame
cv2.imshow('MultiTracker', frame)
outconfig.write(frame)
# -ms 로 지정한 만큼마다, 그 프레임의 데이터와 사진을 저장함.
if framecount % save_image_per_n_milliseconds == 0 :
filedata.writeAndSave(originalframe)
framecount = 0 # int 자료형 overflow 방지
# quit on ESC button
if cv2.waitKey(1) & 0xFF == 27: # Esc pressed
break
print("session end")
# main variable -------------------------------------------------------------
inVideo = cv2.VideoCapture('origin.avi')
success, img = inVideo.read()
height, width = img.shape[:2]
imgSet = [] # set of image for write video
imgCount = 0 # number of processed imgs
avgTime = 0 # to calculate this program's each image processing time
font = cv2.FONT_HERSHEY_SIMPLEX # alarm safety or not
prevDist = 0
GPIO.setmode(GPIO.BOARD)
GPIO.setup(BUZZER_PIN, GPIO.OUT)
buzzer = GPIO.PWM(BUZZER_PIN, 882)
# shoes variable ------------------------------------------------------------
tracker = cv2.MultiTracker_create() # init tracker for tracking shoes
x1, y1, w1, h1 = 720, 870, 940 - 720, 1080 - 870 # first shoe's location
x2, y2, w2, h2 = 1050, 840, 1250 - 1050, 1025 - 840 # second shoe's location
box1 = (x1, y1, w1, h1)
box2 = (x2, y2, w2, h2)
tracker.add(cv2.TrackerBoosting_create(), img,
box1) # start tracking first shoe
tracker.add(cv2.TrackerBoosting_create(), img,
box2) # start tracking second shoe
# stairs variable -----------------------------------------------------------
numOfStairs = 2 # init stair's variables
numOfTester = int(height / 10)
stairInterval = height / numOfStairs
edgeArea = int(stairInterval / numOfStairs) # area of stair's edge
voteStair = [] # for voting stair's edge, save the voting set of stairs
def main(cam1_path, cam2_path):
#Read video object
cam1_video = cv2.VideoCapture(cam1_path)
cam2_video = cv2.VideoCapture(cam2_path)
# Exit if video not opened.
if not cam1_video.isOpened():
print("Could not open cam1_video")
sys.exit()
if not cam2_video.isOpened():
print("Could not open cam2_video")
sys.exit()
#Read first frame.
ok, cam1_image = cam1_video.read()
if not ok:
print('Cannot read cam1_video file')
sys.exit()
ok, cam2_image = cam2_video.read()
if not ok:
print('Cannot read cam2_video file')
sys.exit()
#initialize camera 1, 2 trackers here
cam1_tracker = cv2.MultiTracker_create()
cam2_tracker = cv2.MultiTracker_create()
cam1_personsTracked = []
cam2_personsTracked = []
#cam1_tracker.add(cam1_image, cam1_personsTracked);
#cam2_tracker.add(cam2_image, cam2_personsTracked);
#For each frame call individual trackers and then the centralServer
frameNumber = 0
personDetectionFrameThres = 10
personDetectionFlag = False
while True:
frameNumber = frameNumber + 1
print('*************************************')
print('frameNumber: %d' % frameNumber)
ok, cam1_image = cam1_video.read()
if not ok:
print('cam1_image not read!')
break
ok, cam2_image = cam2_video.read()
if not ok:
print('cam2_image not read!')
break
if (frameNumber == 1 or frameNumber % personDetectionFrameThres == 0):
personDetectionFlag = True
else:
personDetectionFlag = False
#cam1_personsTracked = cam_personTracking(cam1_image, cam1_tracker, personDetectionFlag)#True)#personDetectionFlag)
#cam2_personsTracked = cam_personTracking(cam2_image, cam2_tracker, personDetectionFlag)#True)#personDetectionFlag)
cam1_personsTracked, new_flag_1 = cam_personTracking(
cam1_image, cam1_tracker,
personDetectionFlag) #True)#personDetectionFlag)
cam2_personsTracked, new_flag_2 = cam_personTracking(
cam2_image, cam2_tracker,
personDetectionFlag) #True)#personDetectionFlag)
print('CAMERA 1 INFO: ')
print(cam1_personsTracked)
print('Possible new person added this frame :', new_flag_1)
print('CAMERA 2 INFO: ')
print(cam2_personsTracked)
print('Possible new person added this frame :', new_flag_2)
#aggregatedPersonsList = centralServer(cam1_personsTracked, cam2_personsTracked, cam1_image, cam2_image, new_flag_1, new_flag_2)
aggregatedPersonsList = []
if (len(aggregatedPersonsList)):
#Draw tracks and bbox for individual camera outputs and aggregated outputs
print('-------------------------------------------------------')
print('*************************************')
cam1_image_copy = cam1_image[:]
cam2_image_copy = cam2_image[:]
'''
index = -1
for newbox in cam1_personsTracked:
index = index + 1
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(cam1_image, p1, p2, (255,0,0), 2, 1)
cv2.putText(cam1_image, "ID:"+str(index), p1, cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50,170,50),2);
index = -1
for newbox in cam2_personsTracked:
index = index + 1
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(cam2_image, p1, p2, (255,0,0), 2, 1)
cv2.putText(cam2_image, "ID:"+str(index), p1, cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50,170,50),2);
'''
for person in aggregatedPersonsList:
if (person.vis1):
newbox = person.box1
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(cam1_image_copy, p1, p2, (255, 0, 0), 2, 1)
cv2.putText(cam1_image_copy, "ID:" + str(person.ID), p1,
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
if (person.vis2):
newbox = person.box2
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(cam2_image_copy, p1, p2, (255, 0, 0), 2, 1)
cv2.putText(cam2_image_copy, "ID:" + str(person.ID), p1,
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
cv2.imshow("Camera1", cam1_image)
cv2.imshow("Camera2", cam2_image)
cv2.imshow("Camera1_Central", cam1_image_copy)
cv2.imshow("Camera2_Central", cam2_image_copy)
keyPressed = cv2.waitKey(10)
if (keyPressed == 27):
break
def __init__(self, object_registry):
self._object_registry = object_registry
self._tracker_created = set()
self._multitracker = cv2.MultiTracker_create()
def __init__(self, parent=None):
super().__init__(parent)
self.resize(640, 360)
self.tracking = 0
self.motion_detector = MotionDetector()
self.multiTracker = cv2.MultiTracker_create()
def __init__(self, initial_bboxes, initial_frame):
self.multiTracker = cv2.MultiTracker_create()
formated_bboxes = voc_bboxes_to_coco(initial_bboxes.copy())
for bbox in formated_bboxes:
self.multiTracker.add(cv2.TrackerMedianFlow_create(),
initial_frame, bbox)
def yolo():
parser = argparse.ArgumentParser()
parser.add_argument('-m',
'--model-path',
type=str,
default='./yolov3-coco/',
help='The directory where the model weights and \
configuration files are.')
parser.add_argument('-w',
'--weights',
type=str,
default='./yolov3-coco/yolov3.weights',
help='Path to the file which contains the weights \
for YOLOv3.')
parser.add_argument(
'-cfg',
'--config',
type=str,
default='./yolov3-coco/yolov3.cfg',
help='Path to the configuration file for the YOLOv3 model.')
parser.add_argument('-vo',
'--video-output-path',
type=str,
default='./output.avi',
help='The path of the output video file')
parser.add_argument('-l',
'--labels',
type=str,
default='./yolov3-coco/coco-labels',
help='Path to the file having the \
labels in a new-line seperated way.')
parser.add_argument('-c',
'--confidence',
type=float,
default=0.5,
help='The model will reject boundaries which has a \
probabiity less than the confidence value. \
default: 0.5')
parser.add_argument('-th',
'--threshold',
type=float,
default=0.3,
help='The threshold to use when applying the \
Non-Max Suppresion')
parser.add_argument(
'--download-model',
type=bool,
default=False,
help='Set to True, if the model weights and configurations \
are not present on your local machine.')
parser.add_argument('-t',
'--show-time',
type=bool,
default=False,
help='Show the time taken to infer each image.')
FLAGS, unparsed = parser.parse_known_args()
#print(FLAGS)
# Get the labels
labels = open(FLAGS.labels).read().strip().split('\n')
# Intializing colors to represent each label uniquely
colors = np.random.randint(0, 255, size=(len(labels), 3), dtype='uint8')
# Load the weights and configutation to form the pretrained YOLOv3 model
net = cv.dnn.readNetFromDarknet(FLAGS.config, FLAGS.weights)
# Get the output layer names of the model
layer_names = net.getLayerNames()
layer_names = [
layer_names[i[0] - 1] for i in net.getUnconnectedOutLayers()
]
################################
height, width = frame.shape[:2]
img, bboxes, _, classid, _ = infer_image(net, layer_names, height, width,
frame, colors, labels, FLAGS)
boxes = [] #It's a list now
j = 0
for i in classid:
if i == 0:
print("persons bounding box is: ", bboxes[j])
boxes.append(bboxes[j].copy())
#print(boxes[i])
j = j + 1
print(boxes) #all faces
############################temp ###########33
#for index,value in enumerate(boxes):
global itr
for i in range(len(boxes)):
itr = itr + 1
# Matching part
labels = {}
#Matching part end
y = boxes[i][1]
x = boxes[i][0]
h = boxes[i][3]
w = boxes[i][2]
crop_img = img[y:y + h, x:x + w]
#cv.imwrite(name,crop_img)
detector = MTCNN()
print("I am a detector phewww !")
print(detector.detect_faces(crop_img))
face_cropped = detector.detect_faces(crop_img)
if (len(face_cropped) > 0):
boxes_face = (face_cropped[0]['box'])
y1 = boxes_face[1]
x1 = boxes_face[0]
h1 = boxes_face[3]
w1 = boxes_face[2]
crop_img_2 = crop_img[y1:y1 + h1, x1:x1 + w1]
name = 'dataset/' + str("face") + str(itr) + '.jpg'
#cv.imwrite(name,crop_img_2)
#Matching Part
path = 'dataset/Face' + str(itr)
os.mkdir(path)
name = 'dataset/Face' + str(itr) + '/' + str("person") + str(
itr) + ".jpg"
cv.imwrite(name, crop_img)
name = 'dataset/Face' + str(itr) + '/' + str("face") + str(
itr) + ".jpg"
try:
cv.imwrite(name, crop_img_2)
except:
pass
#amtching
#Faces_Train.py
os.system('faces_train.py')
#Faces_Train.py
with open("labels.pickle", 'rb') as f:
og_labels = pickle.load(f)
labels = {v: k
for k, v in og_labels.items()
} #talk to me about this ;) (key, value pair conversion)
face_cascade = cv2.CascadeClassifier(
'cascade/data/haarcascade_frontalface_alt2.xml'
) #only front of a face
recognizer = cv2.face.LBPHFaceRecognizer_create()
recognizer.read("trainer.yml")
gray = cv2.cvtColor(crop_img, cv2.COLOR_BGR2GRAY)
faces = face_cascade.detectMultiScale(gray,
scaleFactor=1.5,
minNeighbors=5)
#scaleFactor aghe peeche krke dekho
print("Faces", faces)
for (x, y, w, h) in faces:
print(x, y, w, h)
print("Amir")
roi_gray = gray[y:y + h, x:x + w] #roi -> Region of Interest
roi_color = crop_img[y:y + h, x:x + w]
#Recognition part
id_, conf = recognizer.predict(
roi_gray) #id's and confidence level
print(id_, conf, "moz")
if (conf >= 25 and conf <= 85):
print(id_)
print("he")
print(labels[id_])
font = cv2.FONT_HERSHEY_SIMPLEX
name = labels[id_]
color = (255, 255, 255)
stroke = 2
cv2.putText(frame, name, (x, y), font, 1, color, stroke,
cv2.LINE_AA)
##########################temp done#########33
my_tuple = []
for i in bboxes:
my_tuple.append(tuple(i))
#print(my_tuple)
# Create MultiTracker object
multiTracker = cv2.MultiTracker_create()
# Initialize MultiTracker
colors_multi = []
for bbox in my_tuple:
multiTracker.add(createTrackerByName(trackerType), frame, bbox)
colors_multi.append((randint(64, 255), randint(64,
255), randint(64, 255)))
return multiTracker, colors_multi
print("Please press s to select cell(s) to track")
print("Once you are done, press q to go to next date")
# initialize a dictionary that maps strings to their corresponding
# OpenCV object tracker implementations
OPENCV_OBJECT_TRACKERS = {
"csrt": cv2.TrackerCSRT_create,
"kcf": cv2.TrackerKCF_create,
"boosting": cv2.TrackerBoosting_create,
"mil": cv2.TrackerMIL_create,
"tld": cv2.TrackerTLD_create,
"medianflow": cv2.TrackerMedianFlow_create,
"mosse": cv2.TrackerMOSSE_create,
}
# initialize OpenCV's special multi-object tracker
trackers = cv2.MultiTracker_create()
vs = cv2.VideoCapture(
"C:/Users/CoeFamily/Documents/python_awsradar_gui-master/kbox_new/" +
file)
global numpts
numpts = 0
# loop over frames from the video stream
try:
while vs.isOpened():
# grab the current frame, then handle if we are using a
# VideoStream or VideoCapture object
frame = vs.read()
frame = frame[1]
# check to see if we have reached the end of the stream
R_avg = np.average(region1[:,:,2])
(score, diff) = compare_ssim(ref_region, region, full=True)
scored =0
if score < 0.40 and np.average(region1[:,:,2]) > R_avg + 8:
time.sleep(0.3)
scored= 1
return R_avg,scored
if __name__ == "__main__":
score = 0
not_detect = 0
cam = cv2.VideoCapture('123.mp4')
flag =7
tracker1 = cv2.MultiTracker_create()
count = 1
detectFreq = 10
hog = cv2.HOGDescriptor()
hog.setSVMDetector(cv2.HOGDescriptor_getDefaultPeopleDetector())
colour = (238,130,238)
while(cam.isOpened()):
count -=1
ret, frame = cam.read()
frame = np.asarray(frame)
frame = cv2.resize(frame, (1200,720))
frame1 = frame.copy()
if count == 0:
rects,frame = PedestrianDetection.PedestrianDetection(hog, frame)
if (len(rects) == 0):
video = FileVideoStream(videoPath, resolution=(640, 368))
video.start()
frame = video.read()
frame = cv2.resize(frame, (640, 368))
cv2.namedWindow("Foreground", cv2.WINDOW_AUTOSIZE)
## Select boxes
bboxes = []
colors = []
# Specify the tracker type
trackerType = "CSRT"
# Create MultiTracker object
multiTracker = cv2.MultiTracker_create()
selectRegions(frame)
frameHeight, frameWidth, _ = frame.shape
while video.running():
originalFrame = frame.copy()
# get updated location of objects in subsequent frames
success, positives = multiTracker.update(frame)
# draw tracked objects
for i, newbox in enumerate(positives):
p1 = (int(newbox[0]), int(newbox[1]))
p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3]))
cv2.rectangle(frame, p1, p2, colors[i], 2, 1)
def test(self):
# object tracker
OPENCV_OBJECT_TRACKERS = {
"csrt": cv2.TrackerCSRT_create,
"kcf": cv2.TrackerKCF_create,
"boosting": cv2.TrackerBoosting_create,
"mil": cv2.TrackerMIL_create,
"tld": cv2.TrackerTLD_create,
"medianflow": cv2.TrackerMedianFlow_create,
"mosse": cv2.TrackerMOSSE_create
}
initBB = False
# grab the appropriate object tracker using our dictionary of
# video capture
capture = cv2.VideoCapture(self.clip_path)
fps = capture.get(cv2.CAP_PROP_FPS)
# used to deal with frame rate corrections
frame_gap = int(round(fps / self.config.target_fps))
frame_num = 0
frame_height = int(capture.get(4))
frame_width = int(capture.get(3))
fourcc = cv2.VideoWriter_fourcc(*'XVID')
out = cv2.VideoWriter('IR_example.avi', fourcc, fps,
(frame_width, frame_height))
# initialize the frame fifo
curr_fifo = VideoFIFO(self.config, self.config.n_timesteps,
frame_width, frame_height)
# initialize OpenCV's special multi-object tracker
trackers = cv2.MultiTracker_create()
detections = []
predictions = []
fusions = []
flag = 1
# read video frames
while True:
flag, frame = capture.read()
# end of movie
if flag == 0:
break
# use FASTER for detection as long as the tracker is not initialized
if not initBB:
# run faster to detect region every frame
res = predict(self.faster_pred, frame)
# get segments block of previous 13 frames - bb interpolation
human_bb = filter_bb(res)
human_bb = [0, 0, 0, 0]
# skip if no detection
if np.all([x == 0 for x in human_bb]):
# select the bounding box of the object we want to track (make
# sure you press ENTER or SPACE after selecting the ROI)
human_bb = cv2.selectROI("Frame",
frame,
fromCenter=False,
showCrosshair=True)
# OpenCV object tracker objects
tracker = OPENCV_OBJECT_TRACKERS["csrt"]()
# start OpenCV object tracker using the supplied bounding box
trackers.add(tracker, frame, human_bb)
human_detection_det = (human_bb[0], human_bb[1],
human_bb[2] + human_bb[0],
human_bb[3] + human_bb[1])
detections.append(copy.copy(human_detection_det))
predictions.append(0)
fusions.append(0)
else:
for det in human_bb:
# OpenCV object tracker objects
tracker = OPENCV_OBJECT_TRACKERS["csrt"]()
# start OpenCV object tracker using the supplied bounding box
# coordinates, then start the FPS throughput estimator as well
human_detection_tup = (det[0], det[1], det[2] - det[0],
det[3] - det[1])
trackers.add(tracker, frame, human_detection_tup)
detections.append(copy.copy(det))
predictions.append(0)
fusions.append(0)
initBB = True
else:
(success, human_detection_tups) = trackers.update(frame)
for i, human_detection_tuple in enumerate(
human_detection_tups):
# grab the new bounding box coordinates of the object
(x, y, w, h) = [int(v) for v in human_detection_tuple]
human_detection = [x, y, x + w, y + h]
detections[i] = human_detection
if frame_num % frame_gap == 0:
# add to fifo according to the correct frame rate
curr_fifo.add_frame(frame)
for i, human_det in enumerate(detections):
curr_block = curr_fifo.recover_block(
human_det, self.config.new_frame_size[0],
self.config.new_frame_size[1], frame_num)
# run model to get predictions
predictions_add, predictions_mul, fus_mul = self.test_step(
curr_block)
predictions[i] = predictions_mul
fusions[i] = fus_mul
# draw predictions
print(self.label_dict_inv[int(predictions_add)])
print(self.label_dict_inv[int(predictions_mul)])
frame_rec = frame.copy()
for i, h in enumerate(detections):
cv2.rectangle(frame_rec, (h[0], h[1]), (h[2], h[3]),
color=(0, 0, 255),
thickness=1)
cv2.putText(frame_rec,
self.label_dict_inv[int(predictions[i])],
(h[0], h[1]), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(0, 0, 255), 1)
probs1 = self.label_dict_inv[0] + '=' + str(fusions[i][0][0]) + ',' + \
self.label_dict_inv[1] + '=' + str(fusions[i][0][1]) + ',' + \
self.label_dict_inv[2] + '=' + str(fusions[i][0][2])
probs2 = self.label_dict_inv[3] + '=' + str(fusions[i][0][3]) + ',' +\
self.label_dict_inv[4] + '=' + str(fusions[i][0][4])
cv2.putText(frame_rec, probs1, (10, 40),
cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 255), 1)
cv2.putText(frame_rec, probs2, (10, 50),
cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 255), 1)
out.write(frame_rec)
# plot frame
cv2.imshow('vid', frame_rec)
cv2.waitKey(100)
#cv2.destroyAllWindows()
frame_num += 1
# When everything done, release the capture
capture.release()
out.release()
cv2.destroyAllWindows()