def initialize_tracker(i):
tracker_types = [
'BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'MOSSE', 'CSRT'
]
tracker_type = tracker_types[i]
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'CSRT':
tracker = cv2.TrackerCSRT_create()
if tracker_type == 'MOSSE':
tracker = cv2.TrackerMOSSE_create()
return tracker
def initTracker(self, firstFrame, mode=Mode.NO_DETECTION):
if self.selectedTracker == self.trackerTypes[0]:
self.tracker = cv2.TrackerBoosting_create()
elif self.selectedTracker == self.trackerTypes[1]:
self.tracker = cv2.TrackerMIL_create()
elif self.selectedTracker == self.trackerTypes[2]:
self.tracker = cv2.TrackerKCF_create()
elif self.selectedTracker == self.trackerTypes[3]:
self.tracker = cv2.TrackerTLD_create()
elif self.selectedTracker == self.trackerTypes[4]:
self.tracker = cv2.TrackerMedianFlow_create()
elif self.selectedTracker == self.trackerTypes[5]:
self.tracker = cv2.TrackerGOTURN_create()
numpyImage = None
if mode==Mode.DETECTION:
numpyImage = self.detect(firstFrame)
elif mode==Mode.NO_DETECTION:
numpyImage = self.selectBB(firstFrame)
return numpyImage
def call_tracker_constructor(self, tracker_type):
# -- TODO: remove this if I assume OpenCV version > 3.4.0
if int(self.major_ver == 3) and int(self.minor_ver) < 3:
tracker = cv2.Tracker_create(tracker_type)
# --
else:
if tracker_type == 'CSRT':
tracker = cv2.TrackerCSRT_create()
elif tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
elif tracker_type == 'MOSSE':
tracker = cv2.TrackerMOSSE_create()
elif tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
elif tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
elif tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
elif tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
elif tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
return tracker
def createTrackerByName(trackerType):
global trackerTypes
global multiTracker
# 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
return tracker
def tracker_creater(api_name):
# Then create a tracker object for which ever tracker is choosen
if api_name == 'BOOSTING':
api_tracker = cv2.TrackerBoosting_create()
elif api_name == 'MIL':
api_tracker = cv2.TrackerMIL_create()
elif api_name == 'KCF':
api_tracker = cv2.TrackerKCF_create()
elif api_name == 'MEDIANFLOW':
api_tracker = cv2.TrackerMedianFlow_create()
elif api_name == "CSRT":
api_tracker = cv2.TrackerCSRT_create()
elif api_name == "MOSSE":
api_tracker = cv2.TrackerMOSSE_create()
else:
api_tracker = None
print('Incorrect tracker name')
print('Available trackers are:')
for t in tracker_types:
print(t)
return api_tracker
def createOneTracker(type):
(major_ver, minor_ver, subminor_ver) = (cv2.__version__).split('.')
tracker_type = type
if int(minor_ver) < 3:
tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
if tracker_type == 'MOSSE':
tracker = cv2.TrackerMOSSE_create()
if tracker_type == "CSRT":
tracker = cv2.TrackerCSRT_create()
return tracker
def main():
output_folder = 'results_70'
data_folder = "data"
for path, subfolders, files in os.walk(data_folder):
print(path)
if "groundtruth.txt" not in files:
continue
dataset = path.split("\\")[-1]
path_gt = f'{path}/groundtruth.txt'
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(),
}
gt = read_ground_truth(path_gt)
# visualize_bounding_boxes(path, gt)
# continue
for tracker_name, tracker in trackers.items():
print("Computing tracking with", tracker_name)
start = time.time()
predictions = track(path, init_tracker(gt), tracker)
accuracy, robustness = evaluate(predictions, gt, iou_cutoff=0.7)
duration = time.time() - start
print(f'accuracy = {accuracy}, robustness = {robustness}')
f = open(f'{output_folder}/result.txt', 'a+')
f.write('{};{};{:.2f};{:.2f};{:.3f}\n'.format(
dataset, tracker_name, accuracy * 100.0, robustness * 100.0,
duration))
f.close()
f = open(f'{output_folder}/{dataset}_{tracker_name}.txt', 'a+')
for bb in predictions:
f.write(f'{bb}\n')
f.close()
def cam_personTracking(image, tracker, flag):
#print('\tcam_personTracking')
threshold_centroid = 60
th = 40 #boundaryThreshold_centroid
H, W, C = image.shape
new = 0 #number of new persons added
#Do tracking of currently tracked persons
#personsTracked, personsTracked_ID = personTracking(image, tracker) #Main Tracker Update
personsTracked = personTracking(image, tracker) #Main Tracker Update #LIST
#Check for boundary persons and deltracker and create new tracker with valid persons
personsTracked, personRemovedIndexList, new_tracker = cam_boundaryCheck(
image, personsTracked, tracker, th)
#print('\t\tpersonsTracked: %d'%len(personsTracked))
#print(personsTracked)
#If flag is True, detect if new persons are present and add a new tracker for those persons
personsMod = personsTracked #Tuple now but later becomes List
if (flag): #Every tenth frame
personsDetected = personDetection(
image[th:H - th, th:W - th,
0:C], th) #Every 10th image detect all persons
print('personsDetected')
print(personsDetected)
if personsDetected: #if frame not empty
personsMod, new = modifyTracker(personsTracked, personsDetected,
threshold_centroid, new)
#print('\tAm I a List?: ')
#print(isinstance(personsMod, list))
#print('\t\tpersonsDetected: %d'%len(personsDetected));
#print(personsDetected);
#print('\t\tpersonsMod After: %d'%len(personsMod));
#print(personsMod);
for i in range(len(personsTracked), len(personsMod)):
ok = new_tracker.add(cv2.TrackerMIL_create(), image,
tuple(personsMod[i]))
return personsMod, new, personRemovedIndexList, new_tracker
def createOpencvTracker(self, type, detection):
''' Configures the opencv tracker type. '''
if type == 'kcf':
self.trackers_opencv = {
key: cv2.TrackerKCF_create()
for key in range(len(detection))
}
elif type == 'boosting':
self.trackers_opencv = {
key: cv2.TrackerBoosting_create()
for key in range(len(detection))
}
elif type == 'mil':
self.trackers_opencv = {
key: cv2.TrackerMIL_create()
for key in range(len(detection))
}
elif type == 'tld':
self.trackers_opencv = {
key: cv2.TrackerTLD_create()
for key in range(len(detection))
}
elif type == 'medianflow':
self.trackers_opencv = {
key: cv2.TrackerMedianFlow_create()
for key in range(len(detection))
}
elif type == 'csrt':
self.trackers_opencv = {
key: cv2.TrackerCSRT_create()
for key in range(len(detection))
}
elif type == 'mosse':
self.trackers_opencv = {
key: cv2.TrackerMOSSE_create()
for key in range(len(detection))
}
def __init__(self, windowName = 'default window', videoName = "default video"):
self.selection = None # 框选追踪目标状态
self.track_window = None # 追踪窗口状态
self.drag_start = None # 鼠标拖动状态
self.speed = 50 # 视频播放速度
self.video_size = (960,540) # 视频大小
self.box_color = (255,255,255) # 跟踪器外框颜色
self.path_color = (0,0,255) # 路径颜色
# 选择追踪器类型
# 0 1 2 3 4 5 6 7 8
self.tracker_types = ['BOOSTING', 'MIL','KCF', 'TLD', 'MEDIANFLOW', 'GOTURN', 'Dlib_Tracker', 'CamShift','Template_Matching']
self.tracker_type = self.tracker_types[2]
# 创建视频窗口
cv2.namedWindow(windowName,cv2.WINDOW_AUTOSIZE)
cv2.setMouseCallback(windowName,self.onmouse)
self.windowName = windowName
# 打开视频
self.cap = cv2.VideoCapture(videoName)
if not self.cap.isOpened():
print("Video doesn't exit!", videoName)
# 视频相关属性
self.frames_count = int(self.cap.get(cv2.CAP_PROP_FRAME_COUNT)) # 视频总帧数
self.points = deque(maxlen = self.frames_count) # 存放每一帧中追踪目标的中心点
# 判定所选追踪器类型并初始化追踪器
if self.tracker_type == 'BOOSTING':
self.tracker = cv2.TrackerBoosting_create()
elif self.tracker_type == 'MIL':
self.tracker = cv2.TrackerMIL_create()
elif self.tracker_type == 'KCF':
self.tracker = cv2.TrackerKCF_create()
elif self.tracker_type == 'TLD':
self.tracker = cv2.TrackerTLD_create()
elif self.tracker_type == 'MEDIANFLOW':
self.tracker = cv2.TrackerMedianFlow_create()
elif self.tracker_type == 'GOTURN':
self.tracker = cv2.TrackerGOTURN_create()
def create_tracker(tracker: str):
""" Créer un tracker """
# type de tracker selon la précision que l'on veut obtenir ou les fps
tracker_types = [
'BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN', 'MOSSE',
'CSRT'
]
tracker_type = tracker_types[0]
if tracker not in tracker_types:
print(
f'error: not supported tracker \'{tracker}\', available trackers are {tracker_types}.'
)
sys.exit(-1)
if int(major_ver) == 3 and int(minor_ver) < 3:
tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
elif tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
elif tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
elif tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
elif tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
elif tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
elif tracker_type == 'MOSSE':
tracker = cv2.TrackerMOSSE_create()
elif tracker_type == 'CSRT':
tracker = cv2.TrackerCSRT_create()
return tracker
def reset(frame):
global tracker
global multiTracker
multiTracker = None
tracker = None
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
if tracker_type == 'CSRT':
tracker = cv2.TrackerCSRT_create()
if tracker_type=='MOSSE':
tracker = cv2.TrackerMOSSE_create()
bboxes = select2(frame)
multiTracker = createMultitracker(bboxes, frame)
def initTracker():
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
elif tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
elif tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
elif tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
elif tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
elif tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
elif tracker_type == "CSRT":
tracker = cv2.TrackerCSRT_create()
elif tracker_type == "MOSSE":
tracker = cv2.TrackerMOSSE_create()
else:
tracker = None
print('Incorrect tracker name')
print('Available trackers are:')
for t in tracker_types:
print(t)
return tracker
def getTracker(tracker_type=''):
# tracker_types = ['BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN', 'MOSSE']
# tracker_type = tracker_types[2]
import pprint
pprint.pprint(dir(cv2))
if int(minor_ver) < 3 or not tracker_type:
tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
if tracker_type == 'MOSSE':
tracker = cv2.TrackerMOSSE_create()
return tracker
def __init__(self, tracker_type="BOOSTING", draw_coord=True):
# version comparision, some API changed.
(major_ver, minor_ver, subminor_ver) = (cv2.__version__).split('.')
self.tracker_types = [
'BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN'
]
self.tracker_type = tracker_type
self.isWorking = False
self.draw_coord = draw_coord
if int(minor_ver) < 3:
self.tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'BOOSTING':
self.tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
self.tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
self.tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
self.tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
self.tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
self.tracker = cv2.TrackerGOTURN_create()
def __init__(self, Frame, Bbox, Tracker):
""" Initialise the tracker using the bounding box of the object in the frame """
# create tracker
# KCF is the default
tracker = cv2.TrackerKCF_create()
if Tracker is 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
elif Tracker is 'MIL':
tracker = cv2.TrackerMIL_create()
elif Tracker is 'TLD':
tracker = cv2.TrackerTLD_create()
elif Tracker is 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
# initialise the tracker with the source frame
ok = tracker.init(Frame, Bbox)
if not ok:
print("INFO: Object: Failed to initialise object tracker")
self._bbox = Bbox
self._tracker = tracker
def createTrackerByName(self, trackerType):
# Create a tracker based on tracker name
if trackerType == self.trackerTypes[0]:
tracker = cv2.TrackerBoosting_create()
elif trackerType == self.trackerTypes[1]:
tracker = cv2.TrackerMIL_create()
elif trackerType == self.trackerTypes[2]:
tracker = cv2.TrackerKCF_create()
elif trackerType == self.trackerTypes[3]:
tracker = cv2.TrackerTLD_create()
elif trackerType == self.trackerTypes[4]:
tracker = cv2.TrackerMedianFlow_create()
elif trackerType == self.trackerTypes[5]:
tracker = cv2.TrackerMOSSE_create()
elif trackerType == self.trackerTypes[6]:
tracker = cv2.TrackerCSRT_create()
else:
tracker = None
print('Incorrect tracker name')
print('Available trackers are:')
for t in self.trackerTypes:
print(t)
return tracker
def create_tracker():
# Set up tracker.
# Instead of MIL, you can also use
tracker_types = ['BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN']
tracker_type = tracker_types[2]
if int(minor_ver) < 3:
tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
return (tracker, tracker_type)
def get_tracker(tracker_type):
trackers = {
'BOOSTING': (cv2.TrackerBoosting_create(), 2),
'MIL': (cv2.TrackerMIL_create(), 2),
'KCF': (cv2.TrackerKCF_create(), 2),
'TLD': (cv2.TrackerTLD_create(), 2),
'MEDIANFLOW': (cv2.TrackerMedianFlow_create(), 2),
'GOTURN': (cv2.TrackerGOTURN_create(), 2),
'MOSSE': (cv2.TrackerMOSSE_create(), 4),
'CSRT': (cv2.TrackerCSRT_create(), 4)
}
if tracker_type not in trackers:
return None
tracker = trackers[tracker_type]
tracker_min_minor = tracker[1]
if int(MINOR) < tracker_min_minor:
print("OpenCV version don't support")
sys.exit()
tracker = tracker[0]
return tracker
def init_tracker(self, frame, bbox):
tracker_types = [
'BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN'
]
tracker_type = tracker_types[0]
if tracker_type == tracker_types[0]:
self.tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
self.tracker = cv2.TrackerMIL_create()
# if tracker_type == 'KCF':
# self.tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
self.tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
self.tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
self.tracker = cv2.TrackerGOTURN_create()
# Initialize tracker with first frame and bounding box
ok = self.tracker.init(frame, bbox)
self.particle_filter = ParticleFilter()
self.particle_filter = ParticleFilter.init_particles(
self.particle_filter, region=bbox, particlesPerObject=100)
img = frame[int(bbox[1]):int(bbox[1] + bbox[3]),
int(bbox[0]):int(bbox[0] + bbox[2])]
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
self.histo = cv2.calcHist([hsv], [0, 1], None, [180, 256],
[0, 180, 0, 256])
self.points.append(
ParticleFilter.get_particle_center(self.particle_filter))
self.frame_number += 1
return self
def object_tracker(frame_video, x_b, y_b, w_b, h_b, tracker_algo="MDF"):
if tracker_algo == "boosting":
tracker = cv2.TrackerBoosting_create()
if tracker_algo == "CSRT":
tracker = cv2.TrackerCSRT_create()
if tracker_algo == "TLD":
tracker = cv2.TrackerTLD_create()
if tracker_algo == "MIL":
tracker = cv2.TrackerMIL_create()
if tracker_algo == "KCF":
tracker = cv2.TrackerKCF_create()
if tracker_algo == "MDF":
tracker = cv2.TrackerMedianFlow_create()
bbox = (x_b, y_b, w_b, h_b)
tracker.init(frame_video, bbox)
x, y, w, h = x_b, y_b, w_b, h_b
cv2.rectangle(frame_video, (x, y), ((x + w), (y + h)), (0, 255, 0), 3, 3)
cv2.putText(frame_video, "Tracking Started", (100, 75),
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0), 2)
y_cen = (y + y + h) / 2
return frame_video, y_cen
def create_tracker(frame, bbox):
# require use of OpenCV 3.3 or above
#if int(minor_ver) < 3:
# tracker = cv2.Tracker_create(tracker_type)
#else:
# bleh, write better code: tracker_type global
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
ok = tracker.init(frame, bbox)
updates[tracker] = 0
return tracker
if __name__ == '__main__':
# Set up tracker.
# Instead of MIL, you can also use
tracker_types = ['BOOSTING', 'MIL', 'KCF', 'TLD', 'MEDIANFLOW', 'GOTURN']
tracker_type = tracker_types[1]
if int(minor_ver) < 3:
tracker = cv2.Tracker_create(tracker_type)
else:
if tracker_type == 'BOOSTING':
tracker = cv2.TrackerBoosting_create()
if tracker_type == 'MIL':
tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
tracker = cv2.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
tracker = cv2.TrackerMedianFlow_create()
if tracker_type == 'GOTURN':
tracker = cv2.TrackerGOTURN_create()
# Read video
# video = cv2.VideoCapture("/Users/administrator/Downloads/fhd-video.mp4")
video = cv2.VideoCapture(0)
# Exit if video not opened.
if not video.isOpened():
def reset_tracker(self):
self.cv2_tracker = cv2.TrackerMIL_create()
self.bbox = None
self.track_itr = 0
# (major, minor) = cv2.__version__.split(".")[:2]
# print(major, minor)
if not os.path.exists(args.vid_path):
sys.exit("[!] WARNING !! Data(input) path does NOT exist!!")
if (args.vid_path.endswith("mp4") or args.vid_path.endswith("avi")) and args.input_type == 0:
sys.exit("[!] WARNING !! Video file provided but processing mode is IMAGES")
if args.vid_path.endswith("jpg") and args.input_type == 1:
sys.exit("[!] WARNING !! Image file provided but processing mode is VIDEO")
OPENCV_OBJECT_TRACKERS = {
"csrt": cv2.TrackerKCF_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()
}
if args.multi == False:
if args.model == 0:
tracker = OPENCV_OBJECT_TRACKERS[args.tracker]
elif args.model == 1:
tracker = SiamRPNvot()
tracker.load_state_dict(torch.load(join(realpath(dirname(__file__)), './siam_tracker/SiamRPNVOT.model')))
tracker.eval()
else:
sys.exit("[!] Incorrect Model argument: Choose 0 for CV2 tracker and 1 for DaSiamRPN tracker")
else:
def tracker_MIL():
return cv2.TrackerMIL_create()
new_box.append(0)
tracker_ok.append(0)
p1.append(0)
p2.append(0)
centre.append(0)
tracker.append(0)
i += 1
while cap.isOpened():
ok, image = cap.read()
#image=cv2.flip(image,1)
k = cv2.waitKey(1) & 0xff
if k == ord('p'):
if flag < max_trackers:
tracker[flag] = cv2.TrackerMIL_create()
b_box[flag] = cv2.selectROI("Select ROI", image)
tracker_ok[flag] = tracker[flag].init(image, b_box[flag])
flag += 1
#print flag
i = 0
while (i < flag):
tracker_ok[i], new_box[i] = tracker[i].update(image)
if tracker_ok[i]:
p1[i] = (int(new_box[i][0]), int(new_box[i][1]))
p2[i] = (int(new_box[i][0] + new_box[i][2]),
int(new_box[i][1] + new_box[i][3]))
#cv2.rectangle(image, p1, p2, (0,0,200),2)
centre[i] = (int(
print "c1 : ", str(c1)
print "c2 : ", str(c2)
people_count = people_count + nc.crossed(c1, c2, linep1, linep2)
text = "people count: " + str(people_count)
print text
cv.putText(image,
text, (0, 15),
cv.FONT_HERSHEY_SIMPLEX,
0.5, (255, 0, 0),
lineType=cv.LINE_AA)
# #___________________________________________________________________________
boxesHistory = copy.deepcopy(trackerObjects)
#___________________________________________________________________________
[
tracker.add(
cv.TrackerMIL_create(), image,
tuple(
nc.convert2ROI(
[item['topleft']['x'], item['topleft']['y']],
[item['bottomright']['x'], item['bottomright']['y']])))
for item in newTrackerObjects
]
ok, boxes = tracker.update(image)
for iter in range(len(boxes)):
newlocation = nc.convert2Rect(boxes[iter])
trackerObjects[iter]['topleft']['x'] = int(newlocation['topleft']['x'])
trackerObjects[iter]['topleft']['y'] = int(newlocation['topleft']['y'])
trackerObjects[iter]['bottomright']['x'] = int(
newlocation['bottomright']['x'])
trackerObjects[iter]['bottomright']['y'] = int(
newlocation['bottomright']['y'])
def frameProcessor(self, frame, gaussianBlur, useBlank, pixelThreshold,
dilation, averageDifferenceFrames, trackerBool):
"""
This function processes the grabbed frames in 'self.grabbedFrames' using
different techniques.
"""
# A frame will converted to a grayscale image
grayImage = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# Makes sure the first screen is black
if self.referenceFrame is None:
# Initialize initial position of the objects to be tracked
if trackerBool:
for object in range(0, len(self.allROI)):
cutter = VideoCutter(frame)
cutter.windowName = "Select object in chamber " + str(
object + 1)
rect = cutter.selectROI()
self.trackedBoxes.append(rect)
tracker = cv2.TrackerMIL_create()
tracker.init(frame, rect)
self.trackers.append(tracker)
self.referenceFrame = grayImage
updates = []
counter = 0
for position in self.trackers:
success, updatedBox = position.update(frame)
if success:
updates.append(updatedBox)
else:
updates.append(self.trackedBoxes[counter])
counter += 1
self.trackedBoxes = updates
# Max a black and white view of the absolute difference between the
# image and the reference frame
difference = cv2.absdiff(self.referenceFrame, grayImage)
# If a user has chosen the option of 'averaging between difference frames'
if averageDifferenceFrames:
if self.lastDifferenceFrame is None:
self.lastDifferenceFrame = difference
# The average is half the value of every pixel of a difference frame +
# the other half from the next difference frame (sliding window)
image = 0.5 * self.lastDifferenceFrame + 0.5 * difference
else:
image = difference
if gaussianBlur[0]:
# A Gaussian blur will be applied to that frame to cancel out noise
image = cv2.GaussianBlur(image, (gaussianBlur[1], gaussianBlur[1]),
0)
# Threshold the data using the data the user has entered
image = cv2.threshold(image, pixelThreshold, 255, cv2.THRESH_TOZERO)[1]
# Dilate the thresholded image to fill in holes
image = cv2.dilate(image, None, iterations=dilation)
image *= 2
self.lastDifferenceFrame = difference
self.referenceFrame = grayImage
i = 0
# Every ROI is cut out of the processed frame
for ROI in self.allROI:
r = ROI
processedFrame = image[int(r[1]):int(r[1] + r[3]),
int(r[0]):int(r[0] + r[2])]
# The preview is showed if desired
if self.preview:
for box in self.trackedBoxes:
xmid = box[0] + (box[2] / 2)
ymid = box[1] + (box[3] / 2)
cv2.imshow('Preview', processedFrame)
cv2.waitKey(1)
time.sleep(1 / self.fps)
else:
# x is the timestamp/frametime
timestamp = len(self.processedFrames[i][1]) * (1 / self.fps)
self.processedFrames[i][0].append(round(timestamp, 4))
# y is the amount of white pixels in the processed frame
self.processedFrames[i][1].append(np.sum(processedFrame > 0))
i += 1
return
def main():
#Cargar archivos
sistema = platform.system()
if sistema == 'Linux':
path_signs = "/home/pzampella/NimSet/CNN/Hands/Keras/CNN_87_1534855126/"
json_file = open(path_signs + 'model.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
classifier_hands = model_from_json(loaded_model_json)
classifier_hands.load_weights(path_signs + '/model.h5')
clasificador_rostro = cv2.CascadeClassifier("/home/pzampella/opencv-3.4.1/data/lbpcascades/lbpcascade_frontalface_improved.xml") # Cargar el clasificador de rostros
#clasificador_mano = cv2.CascadeClassifier("/home/pzampella/opencv-3.4.1/data/haarcascades/haarcascade_hands.xml") # Cargar el clasificador de_manos
if sistema == 'Windows':
path_signs = r'CNN'
json_file = open(path_signs + r'\model.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
classifier_hands = model_from_json(loaded_model_json)
classifier_hands.load_weights(path_signs + '\model.h5')
clasificador_rostro = cv2.CascadeClassifier("lbpcascade_frontalface_improved.xml") # Cargar el clasificador de rostros
#clasificador_mano = cv2.CascadeClassifier("/home/pzampella/opencv-3.4.1/data/haarcascades/haarcascade_hands.xml") # Cargar el clasificador de_manos
# load weights into new model
classifier_hands.compile(loss='binary_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
print("Loaded model hands from disk")
Path_imagenes = "/home/pzampella/Imagenes/"
if os.path.exists(Path_imagenes):
shutil.rmtree(Path_imagenes)
if not os.path.exists(Path_imagenes):
os.makedirs(Path_imagenes)
min_area = 3000 # Minima area detectable (en pixeles)
piel_min = 0.5 # Porcentaje de piel minimo detectable
bajo = np.array([0, 48, 80], dtype="uint8") # Definir rango minimo de tez de piel
alto = np.array([20, 255, 255], dtype="uint8") # Definir rango maximo de tez de piel
trackers = cv2.MultiTracker_create() # Inicializar tracker
bboxs = [[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]] # Declarar bounding box
print("\n Set the maximum number of hands to be detected. The greater the number, the lower the performance.\n")
max_hands = int(input("Maximum number of hands: "))
type(max_hands)
Webcam = cv2.VideoCapture(0) # Declarar variable de captura de video y almacenar un frame
if Webcam.isOpened() == False: # Revisar si hay acceso a la webcam:
print("Error: Camara no disponible\n\n") # Si no, mostrar error
os.system("pause")
return # Terminar programa y salir
# fin if
FrameLeido, ImgOriginal = Webcam.read()
if not FrameLeido or ImgOriginal is None: # Si el frame no fue leido exitosamente:
print("Error: no se ha podido leer la imagen de la camara\n") # Mostrar error
os.system("pause")
return
# fin if
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (11, 11)) # Generar matriz de Kernel
ImgHSV = cv2.cvtColor(ImgOriginal, cv2.COLOR_BGR2HSV) # Pasar imagen a HSV
Mascara = cv2.inRange(ImgHSV, bajo, alto) # Generar mascara
#Mascara = cv2.erode(Mascara, kernel, iterations=2) # Erosionar mascara
Mascara = cv2.dilate(Mascara, kernel, iterations=3) # Dilatar mascara
Borroso = cv2.GaussianBlur(Mascara, (3, 3), 0) # Difuminar mascara
Imagen = cv2.threshold(Borroso, 60, 255, cv2.THRESH_BINARY)[1] # Pasar mascara a binario
contours, hierarchy = cv2.findContours(Imagen, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) # Extraer contornos
areas = [cv2.contourArea(c) for c in contours]
contador = 0
for i in range(0, len(areas)): # Contar el numero de areas mayores que el minimo
if areas[i] > min_area:
contador += 1
# fin if
# fin for
maximo = min(contador, max_hands) # Definir el numero de areas (maximo 3, manos y cara)
for i in range(0, maximo): # Para cada area
areas = [cv2.contourArea(c) for c in contours] # Extraer areas
max_index = np.argmax(areas) #Hallar el area de mayor tamano
idx = contours[max_index]
x, y, w, h = cv2.boundingRect(idx) # Crear el bounding box que encierra la mayor area
bboxs[i] = (x, y, w, h)
contours.pop(max_index) # Eliminar la mayor area
#fin for
for i in range(0, maximo):
new_tracker = cv2.TrackerMIL_create()
_ = trackers.add(new_tracker, Imagen, bboxs[i]) # Inicializar trackers
# fin for
contador_viejo = maximo
reset_tracker = 0
cuenta = 0
texto = ""
cara_cuenta = 0
fgbg = cv2.createBackgroundSubtractorMOG2()
while cv2.waitKey(1) != 27 and Webcam.isOpened(): # Ejecutar mientras la tecla ESC no sea presionada
cuenta += 1
FrameLeido, ImgOriginal = Webcam.read() # Leer el proximo frame
if not FrameLeido or ImgOriginal is None: # Si el frame no fue leido exitosamente:
print("Error: no se ha podido leer la imagen de la camara\n") # Mostrar error
os.system("pause")
break # Terminar programa y salir
# fin if
ImgHSV = cv2.cvtColor(ImgOriginal, cv2.COLOR_BGR2HSV) # Pasar imagen a HSV
Mascara = cv2.inRange(ImgHSV, bajo, alto) # Generar mascara
#Mascara = cv2.erode(Mascara, kernel, iterations=1) # Erosionar mascara
Mascara = cv2.dilate(Mascara, kernel, iterations=3) # Dilatar mascara
Borroso = cv2.GaussianBlur(Mascara, (3, 3), 0) # Difuminar mascara
Imagen = cv2.threshold(Borroso, 60, 255, cv2.THRESH_BINARY)[1] # Pasar mascara a binario
contours, _ = cv2.findContours(Imagen, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) # Extraer contornos
areas = [cv2.contourArea(c) for c in contours]
contador = 0
for i in range(0, len(areas)): # Contar el numero de areas mayores que el minimo
if areas[i] > min_area:
contador += 1
# fin if
# fin for
###############################################
maximo = min(contador, max_hands) # Definir el numero de areas (maximo 3, manos y cara)
if contador_viejo != maximo or reset_tracker == 1: # Si cambia el numero de objetos o se pierde el tracking de alguno, reconfigurar tracking
trackers.clear()
trackers = cv2.MultiTracker_create() # Reiniciar tracking
for i in range(0, maximo): # Para cada area
areas = [cv2.contourArea(c) for c in contours] # Extraer areas
max_index = np.argmax(areas) # Hallar el area de mayor tamano
idx = contours[max_index]
x, y, w, h = cv2.boundingRect(idx) # Crear el bounding box que encierra la mayor area
bbox = (x, y, w, h)
contours.pop(max_index) # Eliminar la mayor area
new_tracker = cv2.TrackerMIL_create()
_ = trackers.add(new_tracker, Imagen, bbox) # Redefinir el tracker
# fin for
reset_tracker = 0
# fin if
tracking, bboxs = trackers.update(Imagen) # Actualizar tracker
contours, _ = cv2.findContours(Imagen, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
puntos = []
puntos_medios = ""
for bbox in bboxs:
if tracking:
# Tracking exitoso
p = (int(bbox[0] + bbox[2]/2), int(bbox[1] + bbox[3]/2))
puntos_medios = puntos_medios + str(p[0]) + "," + str(p[1])+" "
for i in range(0, len(areas)):
if cv2.pointPolygonTest(contours[i], p, False) == 1.0:
_, _, bbox[2], bbox[3] = cv2.boundingRect(contours[i]) # Actualizar alto y ancho del bounding box
# fin if
# fin for
x_ = int(bbox[0])
y_ = int(bbox[1])
w_ = int(bbox[2])
h_ = int(bbox[3])
if h>w:
alpha = int((h_-w_)/2)
p1 = (x_-alpha, y_)
p2 = (x_+w_+alpha, y_+h_)
else:
alpha = int((w_-h_)/2)
p1 = (x_, y_-alpha)
p2 = (x_+w_, y_+h_+alpha)
tamano1 = len(ImgOriginal)
tamano2 = len(ImgOriginal[0])
if p1[0]<0:
p1=(0, p1[1])
if p1[1]<0:
p1=(p1[0], 0)
if p2[0]>tamano1-1:
p2=(tamano1-1, p2[1])
if p2[1]>tamano2-1:
p2=(p2[0], tamano2-1)
revisar = ImgOriginal[p1[1]:p2[1], p1[0]:p2[0]]
if len(revisar[0]) != 0:
faces = len(clasificador_rostro.detectMultiScale(cv2.cvtColor(revisar, cv2.COLOR_BGR2GRAY), scaleFactor=1.1, minNeighbors=5))
else:
faces = -1
if faces == 0:
puntos.append([p1[1], p2[1], p1[0], p2[0]]) # Calcular puntos del bounding box
cv2.rectangle(ImgOriginal, p1, p2, (200, 0, 0), 2, 1) # Dibujar el nuevo bounding box
#end if
else:
# Tracking fallido
cv2.putText(ImgOriginal, "Fallo durante rastreo", (100, 80), cv2.FONT_HERSHEY_SIMPLEX, 0.75,
(0, 0, 255), 2)
print("Fallo durante rastreo")
# fin if
# fin for
#texto = texto + puntos_medios + "|"
contador_viejo = maximo
for i in range(0, len(puntos)):
piel = 0
if puntos[i][0] != 0 or puntos[i][1] != 0 or puntos[i][2] != 0 or puntos[i][3] != 0: # Si el cuadro uno tiene imagen
Cuadro = Imagen[puntos[i][0]:puntos[i][1], puntos[i][2]:puntos[i][3]] # Recortar imagen
unos = list(Cuadro[0]).count(1)
total = len(Cuadro) * len(Cuadro[0])
if total == 0:
piel = 0
else:
piel = unos / total # Calcular el porcentaje de piel detectada
# fin if
if piel < piel_min:
reset_tracker = 1
# fin if
if (puntos[i][1]-puntos[i][0]>5) and (puntos[i][3]-puntos[i][2]>5):
Aux = ImgOriginal[puntos[i][0]:puntos[i][1], puntos[i][2]:puntos[i][3]]
Mask = Imagen[puntos[i][0]:puntos[i][1], puntos[i][2]:puntos[i][3]]
Mask2 = cv2.morphologyEx(Mask, cv2.MORPH_CLOSE, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5, 5)) )
Aux = cv2.bitwise_and(Aux,Aux,mask = Mask2)
Aux2 = cv2.cvtColor(Aux, cv2.COLOR_RGB2GRAY)
Aux2[Aux2 == 0] = 128
contrast = exposure.equalize_hist(Aux2) * 255
Aux3 = sc.fit_transform(cv2.resize(contrast, (28, 28)))
Aux4 = Aux3.reshape(1, 28, 28, 1)
# plt.imshow(Aux)
# plt.show()
letra = np.argmax(classifier_hands.predict(Aux4))
cv2.putText(ImgOriginal, leer_letra(letra), (puntos[i][2]+5, puntos[i][0]+30), cv2.FONT_HERSHEY_SIMPLEX, 1.0, (0, 0, 255), 2)
# fin if
#cv2.namedWindow("Uno", cv2.WINDOW_NORMAL)
#cv2.imshow("Uno", Aux) # Dibujar el cuadro
#cv2.imwrite(Path_imagenes + str(i) + "_" + str(0) + ".jpg", ImgOriginal[puntos[i][0]:puntos[i][1], puntos[i][2]:puntos[i][3]])
#cv2.imwrite(Path_imagenes + str(i) + "_" + str(cuenta) + ".jpg", ImgOriginal[puntos[i][0]:puntos[i][1], puntos[i][2]:puntos[i][3]]) # Guardar imagen
#fin if
#cv2.namedWindow("Video mascara", cv2.WINDOW_NORMAL)
#cv2.imshow("Video mascara", Aux2)
# fin for
cv2.namedWindow("Video original", cv2.WINDOW_NORMAL) # Crear ventana para mostrar frame original
cv2.imshow("Video original", ImgOriginal) # Mostrar frame original
# fin while
archivo = open(Path_imagenes+"\Coordenadas.txt", "w")
archivo.write(texto)
archivo.close()
return