class videoSession(object):
def __init__(self, videoPath, visualize=False, cnn=False):
self.visualizeFlag = visualize
self.cnnFlag = cnn
self.doFace = False
# Yolo
self.yolo = YOLO()
# Definition of the parameters
max_cosine_distance = 0.3
nn_budget = None
self.nms_max_overlap = 1.0
# deep_sort
model_filename = 'model_data/mars-small128.pb'
self.encoder = gdet.create_box_encoder(model_filename, batch_size=1)
metric = nn_matching.NearestNeighborDistanceMetric(
"cosine", max_cosine_distance, nn_budget)
self.tracker = Tracker(metric)
self.video_capture = filevideostream.FileVideoStream(videoPath)
#cutID
self.cut_id = 0
def start(self):
self.video_capture.start()
def release(self):
self.video_capture.stopprocess()
self.video_capture.stream.release()
def nextFrame(self):
ret, frame_no, frame, isCut = self.video_capture.read()
resDict = {}
if ret != True:
return None
# get yolo boxes
boxs = self.yolo.detect_image(frame)
# print("box_num",len(boxs))
features = self.encoder(frame, boxs)
# score to 1.0 here).
detections = [
Detection(bbox, 1.0, feature)
for bbox, feature in zip(boxs, features)
]
# Run non-maxima suppression.
boxes = np.array([d.tlwh for d in detections])
scores = np.array([d.confidence for d in detections])
indices = preprocessing.non_max_suppression(boxes,
self.nms_max_overlap,
scores)
detections = [detections[i] for i in indices]
# Call the tracker
if isCut:
self.tracker.delete_all()
self.cut_id += 1
self.tracker.predict()
self.tracker.update(detections)
resDict['frame_no'] = frame_no
resDict['is_cut'] = isCut
resDict['cut_id'] = self.cut_id
resDict['person'] = []
for track in self.tracker.tracks:
if not track.is_confirmed() or track.time_since_update > 1:
continue
bbox = track.to_tlbr()
x1 = int(bbox[0]) if int(bbox[0]) >= 0 else 0
y1 = int(bbox[1]) if int(bbox[1]) >= 0 else 0
x2 = int(bbox[2]) if int(bbox[2]) <= 1920 else 1920
y2 = int(bbox[3]) if int(bbox[3]) <= 1080 else 1080
pifDict = {
'trackId': track.track_id,
'x1': x1,
'y1': y1,
'x2': x2,
'y2': y2
}
if (y2 - y1) * (x2 - x1) < 500:
continue
peopleFrame = frame[y1:y2, x1:x2]
faceDict = None
face_locations = []
if self.doFace:
if self.cnnFlag:
try:
face_locations = face_recognition.face_locations(
peopleFrame, model='cnn')
except RuntimeError as e:
print('error!')
print(e)
face_locations = []
else:
face_locations = face_recognition.face_locations(
peopleFrame)
for top, right, bottom, left in face_locations:
faceDict = {'x1': left, 'x2': right, 'y1': top, 'y2': bottom}
break
pifDict['face'] = faceDict
resDict['person'].append(pifDict)
if self.visualizeFlag:
cv2.putText(
frame,
'frame %d, cut %d' % (resDict['frame_no'], resDict['cut_id']),
(0, 20), 0, 1, (0, 255, 0), 2)
for people in resDict['person']:
cv2.rectangle(frame, (people['x1'], people['y1']),
(people["x2"], people['y2']), (255, 255, 255), 2)
if people['face']:
x1 = people['x1'] + people['face']['x1']
x2 = people['x1'] + people['face']['x2']
y1 = people['y1'] + people['face']['y1']
y2 = people['y1'] + people['face']['y2']
cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 0, 255), 2)
cv2.imshow('', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
return
return resDict
def main():
yolo = YOLO()
# Definition of the parameters
max_cosine_distance = 0.3
nn_budget = None
nms_max_overlap = 1.0
# deep_sort
model_filename = 'model_data/mars-small128.pb'
encoder = gdet.create_box_encoder(model_filename, batch_size=1)
metric = nn_matching.NearestNeighborDistanceMetric("cosine",
max_cosine_distance,
nn_budget)
tracker = Tracker(metric)
writeVideo_flag = False
doFace_flag = True
video_capture = filevideostream.FileVideoStream(
"./testVideo/604_0_new.mp4")
video_capture.start()
# cv2.VideoCapture("/media/seb101-user/DATA/TestV_videos/447_1_old.mp4")
#CutDetector
#cutDetector = Util.CutDetectior.CutDetector(threshold=0.3)
if writeVideo_flag:
# Define the codec and create VideoWriter object
w = int(video_capture.get(3))
h = int(video_capture.get(4))
print(w)
print(h)
fourcc = cv2.VideoWriter_fourcc(*'MJPG')
out = cv2.VideoWriter('output.avi', fourcc, 15, (w, h))
list_file = open('detection.txt', 'w')
frame_index = -1
fps = 0.0
#yolo multiprocess queues/values
while True:
tfps = time.time()
tget = time.time()
ret, frame_no, frame, isCut = video_capture.read()
if ret != True:
break
tget = time.time() - tget
# isCut = cutDetector.putFrame(frame)
# people_frame = []
#face_locations = face_recognition.face_locations(frame, model="cnn")
# image = Image.fromarray(frame)
#image = Image.fromarray(frame[...,::-1]) #bgr to rgb
tyolo = time.time()
boxs = yolo.detect_image(frame)
# print("box_num",len(boxs))
features = encoder(frame, boxs)
# score to 1.0 here).
detections = [
Detection(bbox, 1.0, feature)
for bbox, feature in zip(boxs, features)
]
tyolo = time.time() - tyolo
# Run non-maxima suppression.
boxes = np.array([d.tlwh for d in detections])
scores = np.array([d.confidence for d in detections])
indices = preprocessing.non_max_suppression(boxes, nms_max_overlap,
scores)
detections = [detections[i] for i in indices]
# Call the tracker
ttrack = time.time()
if isCut:
tracker.delete_all()
tracker.predict()
tracker.update(detections)
ttrack = time.time() - ttrack
#draw image
tface = time.time()
for det in detections:
bbox = det.to_tlbr()
x1 = int(bbox[0])
y1 = int(bbox[1])
x2 = int(bbox[2])
y2 = int(bbox[3])
peopleFrame = frame[y1:y2, x1:x2]
if doFace_flag:
face_locations = face_recognition.face_locations(peopleFrame,
model='cnn')
for top, right, bottom, left in face_locations:
top += y1
bottom += y1
right += x1
left += x1
cv2.rectangle(frame, (left, top), (right, bottom),
(0, 0, 255), 2)
tface = time.time() - tface
tvis = time.time()
for det in detections:
bbox = det.to_tlbr()
x1 = int(bbox[0])
y1 = int(bbox[1])
x2 = int(bbox[2])
y2 = int(bbox[3])
#cv2.rectangle(frame, (x1, y1), (x2, y2), (255, 0, 0), 2)
for track in tracker.tracks:
if not track.is_confirmed() or track.time_since_update > 1:
continue
bbox = track.to_tlbr()
cv2.rectangle(frame, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])), (255, 255, 255), 2)
cv2.putText(frame, str(track.track_id),
(int(bbox[0]) + 50, int(bbox[1]) + 50), 0, 5e-3 * 200,
(0, 255, 0), 2)
cv2.imshow('', frame)
tvis = time.time() - tvis
if writeVideo_flag:
# save a frame
out.write(frame)
frame_index = frame_index + 1
list_file.write(str(frame_index) + ' ')
if len(boxs) != 0:
for i in range(0, len(boxs)):
list_file.write(
str(boxs[i][0]) + ' ' + str(boxs[i][1]) + ' ' +
str(boxs[i][2]) + ' ' + str(boxs[i][3]) + ' ')
list_file.write('\n')
fps = 1 / (time.time() - tfps)
print(
"fps= %.2f, frame:%0.f, tget:%.2f tyolo:%.2f, ttrack:%.2f, tface:%.2f, tvis: %.2f"
% (fps, frame_no, tget * 1000, tyolo * 1000, ttrack * 1000,
tface * 1000, tvis * 1000))
if isCut:
print(
"=============================================================================================="
)
# Press Q to stop!
if cv2.waitKey(1) & 0xFF == ord('q'):
break
video_capture.stop()
if writeVideo_flag:
out.release()
list_file.close()
cv2.destroyAllWindows()
