about matching detections and trackers

[3073]

hi helper ... can I take the principle you used for matching detection and trackers and use dlib for creating a new tracker for the unmatched detection?

[kcg2015]

Sorry for the late reply. It should be the same principle

[3073]

Thank you as always... normally the detection will run every 30 frames ( some frames will be skipped) for increasing the speed of processing because the detection is slow. anyways i will provide you the result soon for asking a help ...

…

On Sun, Mar 17, 2019 at 6:48 PM Kyle Guan ***@***.***> wrote: Sorry for the late reply. It should be the same principle — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#19 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AgXgdffDDFtBqBLu9qO0WwD1Ep3Hce8Gks5vXvBigaJpZM4b2H67> .

[kcg2015]

Cool!

[3073]

##### there are some code here not necessary for this particular discussion. def iou(self,a, b): ### I used your code directly here w_intsec = np.maximum (0, (np.minimum(a[2], b[2]) -np.maximum(a[0], b[0]))) h_intsec = np.maximum (0, (np.minimum(a[3], b[3]) -np.maximum(a[1], b[1]))) s_intsec = w_intsec * h_intsec s_a = (a[2] - a[0])*(a[3] - a[1]) s_b = (b[2] - b[0])*(b[3] - b[1]) return float(s_intsec)/(s_a + s_b -s_intsec) def assign_detections_to_trackers(self ,trackers, detections, iou_thrd= 0.6): ### I used your code here unmatched_trackers, unmatched_detections = [], [] IOU_mat= np.zeros((len(trackers),len(detections)),dtype=np.float32) for t,trk in enumerate(trackers): for d,det in enumerate(detections): IOU_mat[t,d] = self.iou(trk,det) matched_idx = linear_assignment(-IOU_mat) unmatched_trackers, unmatched_detections = [], [] unmatched_trackers_box, unmatched_detections_box = [], [] for t,trk in enumerate(trackers): if(t not in matched_idx[:,0]): unmatched_trackers.append(t) unmatched_trackers_box.append(trk) for d, det in enumerate(detections): if(d not in matched_idx[:,1]): unmatched_detections.append(d) self.unmatched_detections_box.append(det) matches = [] for m in matched_idx: if(IOU_mat[m[0],m[1]]<iou_thrd): unmatched_trackers.append(m[0]) unmatched_detections.append(m[1]) else: matches.append(m.reshape(1,2)) if(len(matches)==0): matches = np.empty((0,2),dtype=int) else: matches = np.concatenate(matches,axis=0) print(matches, unmatched_detections, unmatched_trackers) return matches, np.array(unmatched_detections), np.array(unmatched_trackers) def people(self, frame): #### start to read the code here self.rects = [] rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) if self.totalFrames % 20 == 0 : ### every 20 frame there will be detection points = self.det.get_points(frame) #### detection model will be executed assignation = self.assign_detections_to_trackers(self.prev_points,points,iou_thrd = 0.6) if len(assignation[0]) > 0: #### updating the tracked objects previously if a match is found tracker_list =[] for trk_idx,det_idx in assignation[0]: tmp_trk = self.trackers[trk_idx] tracker_list.append(tmp_trk) self.update_tracker(tracker_list,frame) if len(assignation[2]) > 0: #### delete the tracker if there is unmatched tracker indexes = [] for trk_idx in assignation[2]: indexes.append(trk_idx) for index in sorted(indexes, reverse=True): del self.trackers[index] if len(assignation[1])> 0: ### generate new tracker for the new object detected dete_points = [] for idx in assignation[1]: point = points[idx] dete_points.append(point) self.generate_tracker(frame,dete_points) else: #### if the frame number is not divisible by 20 updating the trackers will continue self.update_tracker(self.trackers,frame) self.prev_points = self.rects ### previous updated tackers will be noted for calculating the overlap people = self.finder.update(self.rects) ##HERE GOES ANOTHER CODE !!! return frame def update_tracker(self,trackers,frame): for tracker in trackers: tracker.update(frame) pos = tracker.get_position() startX = int(pos.left()) startY = int(pos.top()) endX = int(pos.right()) endY = int(pos.bottom()) rect = (startX, startY, endX, endY) self.rects.append(rect) cv2.rectangle(frame, (startX, startY), (endX, endY), (250,250,250), 4) return self.rects def generate_tracker(self,frame,detections_box): for dete_point in detections_box: (startX, startY, endX, endY) = dete_point tracker = dlib.correlation_tracker() rect = dlib.rectangle(int(startX), int(startY), int(endX), int(endY)) tracker.start_track(frame, rect) self.trackers.append(tracker) pos = tracker.get_position() startX = int(pos.left()) startY = int(pos.top()) endX = int(pos.right()) endY = int(pos.bottom()) rect = (startX, startY, endX, endY) self.rects.append(rect) cv2.rectangle(frame, (startX, startY), (endX, endY), (0,0,250), 4) return self.rects #### this code works fine but i need some help.... ###### Questions to discuss on how do we select the best frame skipped to make it a general purpose code ? for instance for this particular video the best result is obtained when the skipped frame is 20. since this is a counting code a person that did not detected at the detection frame number for example 20 will not be tracked for the next 20 frames ( up to frame number 40). so if the person get lost after the 40 frame , it will not be counted at all and that is a counting error..... do you have a fix to this? will great and honest thanks

[kcg2015]

@3073 , sorry for the late reply. This seems to me a very difficult problem to solve, honestly. I have been struggling with finding parameters that work well with a wide range of scenarios.

[3073]

thank you....

[kcg2015]

@3073 , keep me posted if you find a solution.

[3073]

ok sir

…

On Sun, Mar 24, 2019 at 6:47 PM Kyle Guan ***@***.***> wrote: @3073 <https://github.com/3073> , keep me posted if you find a solution. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#19 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AgXgdeVK1VUJfdvF8YLIK0Jt5emb50K1ks5vaCqVgaJpZM4b2H67> .