def run_detection(self, image, encoder, frame_id):
boxes, confidence, classes = self.detect_image(image)
features = encoder(image, boxes)
detections = [
Detection(bbox, 1.0, cls,
feature) for bbox, _, cls, feature in zip(
boxes, confidence, classes, 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.cfg.DEEPSORT.NMS_MAX_OVERLAP, scores)
detections = [detections[i] for i in indices]
detections_in_ROI = []
print("[INFO] detected: ", len(detections))
for det in detections:
bbox = det.to_tlbr()
centroid_det = (int(
(bbox[0] + bbox[2]) // 2), int((bbox[1] + bbox[3]) // 2))
if check_in_polygon(centroid_det, self.polygon_ROI):
detections_in_ROI.append(det)
print("-----------------")
# return detections_in_ROI
return detections, detections_in_ROI
def read_detection(self, image, frame_info, encoder, frame_id):
detect_folder_path = self.args.read_detect
detect_file_path = os.path.join(detect_folder_path,
frame_info + ".txt")
# file text store path to each frame
detect_file = open(detect_file_path, 'r')
lines = detect_file.readlines()
boxes = []
confidence = []
classes = []
for line in lines:
detect = line.split()
bbox = [
int(detect[0]),
int(detect[1]),
int(detect[2]),
int(detect[3])
]
score = float(detect[4])
class_id = int(detect[5])
boxes.append(bbox)
confidence.append(score)
classes.append(class_id)
features = encoder(image, boxes)
detections = [
Detection(bbox, 1.0, cls,
feature) for bbox, _, cls, feature in zip(
boxes, confidence, classes, 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.cfg.DEEPSORT.NMS_MAX_OVERLAP, scores)
detections = [detections[i] for i in indices]
detections_in_ROI = []
# print("[INFO] detected: ", len(detections))
for det in detections:
bbox = det.to_tlbr()
centroid_det = (int(
(bbox[0] + bbox[2]) // 2), int((bbox[1] + bbox[3]) // 2))
if check_in_polygon(centroid_det, self.polygon_ROI):
detections_in_ROI.append(det)
print("[INFO] detections in ROI: ", len(detections_in_ROI))
print("-----------------")
# return detections_in_ROI
return detections, detections_in_ROI
def find_MOI_candidate(self, region_list, centroid_list):
MOI_candidate = []
for index, region in enumerate(region_list):
region = Polygon(region)
centroids_in_region = [
centroid for centroid in centroid_list
if check_in_polygon(centroid, region)
]
percent_point = len(centroids_in_region) / len(centroid_list)
if percent_point >= self.cfg.CAM.PIM_THRESHOLD:
MOI_candidate.append(index + 1)
return MOI_candidate
def counting(self,
count_frame,
cropped_frame,
_frame,
objs_dict,
counted_obj,
arr_cnt_class,
clf_model=None,
clf_labels=None):
vehicles_detection_list = []
frame_id = count_frame
class_id = None
for (track_id, info_obj) in objs_dict.items():
centroid = info_obj['centroid']
if int(
track_id
) in counted_obj: #check if track_id in counted_object ignore it
continue
#if track_id not in counted object then check if centroid in range of ROI then count it
if len(centroid) != 0 and check_in_polygon(
centroid, self.polygon_ROI
) == False and info_obj['flag_in_out'] == 1:
info_obj['point_out'] = centroid
if self.use_classify:
bbox = info_obj['best_bbox']
obj_img = cropped_frame[
int(bbox[1]):int(bbox[3]),
int(bbox[0]):int(
bbox[2]), :] #crop obj following bbox for clf
class_id = self.run_classifier(clf_model, clf_labels,
obj_img)
if class_id == -1:
continue
else:
class_id = info_obj['class_id']
# MOI of obj
moi, _ = MOI.compute_MOI(self.cfg, info_obj['point_in'],
info_obj['point_out'])
counted_obj.append(int(track_id))
class_id = self.compare_class(class_id)
arr_cnt_class[class_id][moi - 1] += 1
print("[INFO] arr_cnt_class: \n", arr_cnt_class)
vehicles_detection_list.append((frame_id, moi, class_id + 1))
print("--------------")
return _frame, arr_cnt_class, vehicles_detection_list
def run_detection(self, image, encoder, tracking, frame_id):
boxes, confidence, classes = self.detector(image)
features = encoder(image, boxes)
detections = [
Detection(bbox, 1.0, cls,
feature) for bbox, _, cls, feature in zip(
boxes, confidence, classes, 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.cfg.DEEPSORT.NMS_MAX_OVERLAP, scores)
detections = [detections[i] for i in indices]
detections_in_ROI = []
print("[INFO] detected: ", len(detections))
for det in detections:
bbox = det.to_tlbr()
centroid_det = (int(
(bbox[0] + bbox[2]) // 2), int((bbox[1] + bbox[3]) // 2))
if check_in_polygon(centroid_det, self.TRACKING_ROI):
detections_in_ROI.append(det)
print("[INFO] detections in ROI: ", len(detections_in_ROI))
logFile = os.path.join(log_detected_cam_dir,
'frame_' + str(frame_id) + '.txt')
with open(logFile, "a+") as f:
# for det in detections_in_ROI:
for det in detections:
bbox = det.to_tlbr()
score = "%.2f" % round(det.confidence * 100, 2) + "%"
if len(classes) > 0:
cls = det.cls
# write log file
f.write("{} {} {} {} {} {}\n".format(
int(bbox[0]), int(bbox[1]), int(bbox[2]), int(bbox[3]),
round(det.confidence * 100, 2), cls))
print("-----------------")
# return detections_in_ROI
return detections, detections_in_ROI
def counting(self,
count_frame,
cropped_frame,
_frame,
objs_dict,
counted_obj,
arr_cnt_class,
clf_model=None,
clf_labels=None):
vehicles_detection_list = []
frame_id = count_frame
class_id = None
cv2.putText(_frame, "Frame ID: {}".format(str(frame_id)), (1000, 70),
cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 255), 2)
for (track_id, info_obj) in objs_dict.items():
centroid = info_obj['centroid']
if int(
track_id
) in counted_obj: # check if track_id in counted_object ignore it
continue
# if track_id not in counted object then check if centroid in range of ROI then count it
if (info_obj['last_frame'] + 2 < count_frame
and info_obj['flag_in_out'] == 1) or (
check_in_polygon(centroid, self.polygon_ROI) == False
and info_obj['flag_in_out'] == 1):
info_obj['point_out'] = centroid
class_id = self.voting_class(info_obj['class_list'])
# ignore special class not in contest
if class_id == 4:
continue
# compute MOI of obj
if self.count_method == 1:
moi = MOI.compute_MOI_cosine(self.cfg,
info_obj['point_in'],
info_obj['point_out'])
elif self.count_method == 2:
moi, _, _ = MOI.compute_MOI(self.cfg, info_obj['point_in'],
info_obj['point_out'])
elif self.count_method == 3:
moi, _, count = MOI.compute_MOI(self.cfg,
info_obj['point_in'],
info_obj['point_out'])
if count == 0 or count > 1:
moi = MOI.compute_MOI_cosine(self.cfg,
info_obj['point_in'],
info_obj['point_out'])
elif self.count_method == 4:
MOI_candidate = self.find_MOI_candidate(
self.cfg.CAM.ROI_SPLIT_REGION,
info_obj['centroid_list'])
moi, count = MOI.compute_MOI_from_candidate(
self.cfg, info_obj['point_in'], info_obj['point_out'],
MOI_candidate)
if count == 0 or count > 1:
moi = MOI.compute_MOI_cosine_from_candidate(
self.cfg, info_obj['point_in'],
info_obj['point_out'], MOI_candidate)
# mark objs which are counted
counted_obj.append(int(track_id))
if moi > 0:
info_obj['moi'] = moi
track_distance = math.sqrt((info_obj['point_out'][0] -
info_obj['point_in'][0])**2 +
(info_obj['point_out'][1] -
info_obj['point_in'][1])**2)
if track_distance < self.cfg.CAM.D_THRESHOLD[moi - 1]:
continue
if self.args.frame_estimate:
distance_point_line = self.distance_point2roi(
centroid, self.cfg.CAM.LINE_OUT_ROI[moi - 1][0],
self.cfg.CAM.LINE_OUT_ROI[moi - 1][1])
info_obj['frame'] = frame_id + self.estimate_frame(
info_obj['centroid_deque'][0],
info_obj['centroid_deque'][-1], moi,
info_obj['last_bbox'], distance_point_line)
else:
info_obj['frame'] = frame_id
# visualize when obj out the ROI
if info_obj['frame'] == frame_id:
cv2.circle(_frame,
(int(centroid[0]), int(centroid[1])), 12,
self.color_list[moi - 1], -1)
cv2.putText(_frame,
str(class_id + 1) + '.' + str(track_id),
(int(centroid[0]) - 3, int(centroid[1])),
cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(255, 255, 255), 2)
arr_cnt_class[class_id][moi - 1] += 1
vehicles_detection_list.append(
(info_obj['frame'], moi, class_id + 1))
print("--------------")
return _frame, arr_cnt_class, vehicles_detection_list
def draw_tracking(self, image, tracker, tracking, detections, frame_id,
objs_dict):
if tracking:
# Call the tracker
tracker.predict()
tracker.update(detections)
print("[INFO] track in ROI: ", len(tracker.tracks))
for det in detections:
bbox_det = det.to_tlbr()
cv2.rectangle(image, (int(bbox_det[0]), int(bbox_det[1])),
(int(bbox_det[2]), int(bbox_det[3])),
(0, 0, 255), 1)
for track in tracker.tracks:
if not track.is_confirmed() or track.time_since_update > 1:
continue
bbox = track.to_tlbr()
centroid = (int(
(bbox[0] + bbox[2]) // 2), int((bbox[1] + bbox[3]) // 2))
if track.track_id not in objs_dict:
objs_dict.update({
track.track_id: {
'flag_in_out': 0,
'best_bbox': track.det_best_bbox,
'best_bboxconf': track.det_confidence,
'class_id': track.det_class,
'frame': -1,
'centroid_list': [],
'class_list': []
}
}) # frame when vehicle out ROI BTC (frame estimate)
# get the first point(x,y) when obj move into ROI
if len(centroid) != 0 and check_in_polygon(
centroid, self.polygon_ROI) and objs_dict[
track.track_id]['flag_in_out'] == 0:
objs_dict[track.track_id].update({
'flag_in_out': 1,
'point_in': centroid,
'point_out': None,
'frame_in': frame_id
})
# if bbox conf of obj < bbox conf in new frame ==> update best bbox conf
if objs_dict[track.
track_id]['best_bboxconf'] < track.det_confidence:
objs_dict[track.track_id].update({
'best_bbox':
track.det_best_bbox,
'best_bboxconf':
track.det_confidence,
'class_id':
track.det_class
})
objs_dict[track.track_id][
'centroid'] = centroid # update position of obj each frame
objs_dict[track.track_id]['centroid_list'].append(centroid)
objs_dict[track.track_id]['last_bbox'] = bbox
objs_dict[track.track_id]['last_frame'] = frame_id
objs_dict[track.track_id]['class_list'].append(track.det_class)
cv2.rectangle(image, (int(bbox[0]), int(bbox[1]) - 15),
(int(bbox[0] + 50), int(bbox[1])),
(255, 255, 255), -1)
cv2.rectangle(image, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])), (255, 255, 255), 1)
cv2.putText(
image,
str(track.det_class + 1) + "." + str(track.track_id),
(int(bbox[0]), int(bbox[1]) - 1), 0, 0.5, (0, 0, 0), 1)
cv2.circle(image, (centroid[0], centroid[1]), 4, (0, 255, 0),
-1)
# draw track line
image = track.draw_track_line(image)
print("----------------")
return image, objs_dict
def draw_tracking(self, image, tracker, tracking, detections, frame_id,
objs_dict):
if tracking:
# if not check_in_polygon(center, ROI_POLYGON):
# continue
# Call the tracker
tracker.predict()
tracker.update(detections)
print("[INFO] track in ROI: ", len(tracker.tracks))
logFile = os.path.join(log_tracking_cam_dir,
'frame_' + str(frame_id) + '.txt')
with open(logFile, "a+") as f:
for (track, det) in zip(tracker.tracks, detections):
bbox_det = det.to_tlbr()
cv2.rectangle(image, (int(bbox_det[0]), int(bbox_det[1])),
(int(bbox_det[2]), int(bbox_det[3])),
(0, 0, 255), 2)
if not track.is_confirmed() or track.time_since_update > 1:
continue
bbox = track.to_tlbr()
centroid = (int(
(bbox[0] + bbox[2]) // 2), int(
(bbox[1] + bbox[3]) // 2))
if track.track_id not in objs_dict:
objs_dict.update({
track.track_id: {
'flag_in_out': 0,
'best_bbox': det.to_tlbr(),
'best_bboxconf': det.confidence,
'class_id': det.cls
}
})
# get the first point(x,y) when obj move into ROI
if len(centroid) != 0 and check_in_polygon(
centroid, self.polygon_ROI) and objs_dict[
track.track_id]['flag_in_out'] == 0:
objs_dict[track.track_id].update({
'flag_in_out': 1,
'point_in': centroid,
'point_out': None
})
# if bbox conf of obj < bbox conf in new frame ==> update best bbox conf
if objs_dict[
track.track_id]['best_bboxconf'] < det.confidence:
objs_dict[track.track_id].update({
'best_bbox':
det.to_tlbr(),
'best_bboxconf':
det.confidence,
'class_id':
det.cls
})
objs_dict[track.track_id][
'centroid'] = centroid # update position of obj each frame
cv2.rectangle(image, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])),
(255, 255, 255), 2)
cv2.putText(image, "ID: " + str(track.track_id),
(int(bbox[0]), int(bbox[1])), 0,
1e-3 * image.shape[0], (0, 255, 0), 1)
cv2.circle(image, (centroid[0], centroid[1]), 4,
(0, 255, 0), -1)
# draw track line
image = track.draw_track_line(image)
class_name = det.cls
print("[INFO] class_name: ", class_name)
# write log file
f.write("{} {} {} {} {}\n".format(int(bbox[0]),
int(bbox[1]),
int(bbox[2]),
int(bbox[3]),
class_name))
# print("[INFO] dict_MOI after added: ", dict_MOI)
# print("[INFO] dict_detections: after added", dict_detections)
print("----------------")
return image, objs_dict