def update(self, dets, img=None):
"""
Params:
dets - a numpy array of detections in the format [[x,y,w,h,score],[x,y,w,h,score],...]
Requires: this method must be called once for each frame even with empty detections.
Returns the a similar array, where the last column is the object ID.
NOTE: The number of objects returned may differ from the number of detections provided.
"""
self.frame_count += 1
#get predicted locations from existing trackers.
trks = np.zeros((len(self.trackers), 5))
to_del = []
ret = []
for t, trk in enumerate(trks):
pos = self.trackers[t].predict(img) #for kal!
#print(pos)
trk[:] = [pos[0], pos[1], pos[2], pos[3], 0]
if (np.any(np.isnan(pos))):
to_del.append(t)
trks = np.ma.compress_rows(np.ma.masked_invalid(trks))
for t in reversed(to_del):
self.trackers.pop(t)
if dets != []:
matched, unmatched_dets, unmatched_trks = associate_detections_to_trackers(
dets, trks)
#update matched trackers with assigned detections
for t, trk in enumerate(self.trackers):
if (t not in unmatched_trks):
d = matched[np.where(matched[:, 1] == t)[0], 0]
trk.update(dets[d, :][0],
img) ## for dlib re-intialize the trackers ?!
#create and initialise new trackers for unmatched detections
for i in unmatched_dets:
if not self.use_dlib:
trk = KalmanBoxTracker(dets[i, :])
else:
# print(dets[i,:])
trk = CorrelationTracker(dets[i, :], img)
self.trackers.append(trk)
i = len(self.trackers)
for trk in reversed(self.trackers):
if dets == []:
trk.update([], img)
d = trk.get_state()
if ((trk.time_since_update < 1)
and (trk.hit_streak >= self.min_hits
or self.frame_count <= self.min_hits)):
ret.append(np.concatenate((d, [trk.id + 1])).reshape(
1, -1)) # +1 as MOT benchmark requires positive
i -= 1
#remove dead tracklet
if (trk.time_since_update > self.max_age):
self.trackers.pop(i)
if (len(ret) > 0):
return np.concatenate(ret)
return np.empty((0, 5))
def update(self, dets, img_size, root_dic, addtional_attribute_list, predict_num):
"""
Params:
dets - a numpy array of detections in the format [[x,y,w,h,score],[x,y,w,h,score],...]
Requires: this method must be called once for each frame even with empty detections.
Returns the a similar array, where the last column is the object ID.
NOTE:as in practical realtime MOT, the detector doesn't run on every single frame
"""
self.frame_count += 1
# get predicted locations from existing trackers.
trks = np.zeros((len(self.trackers), 5))
to_del = []
ret = []
for t, trk in enumerate(trks):
pos = self.trackers[t].predict() # kalman predict ,very fast ,<1ms
trk[:] = [pos[0], pos[1], pos[2], pos[3], 0]
if np.any(np.isnan(pos)):
to_del.append(t)
trks = np.ma.compress_rows(np.ma.masked_invalid(trks))
for t in reversed(to_del):
self.trackers.pop(t)
if dets != []:
matched, unmatched_dets, unmatched_trks = associate_detections_to_trackers(dets, trks)
# update matched trackers with assigned detections
for t, trk in enumerate(self.trackers):
if t not in unmatched_trks:
d = matched[np.where(matched[:, 1] == t)[0], 0]
trk.update(dets[d, :][0])
trk.face_addtional_attribute.append(addtional_attribute_list[d[0]])
# create and initialise new trackers for unmatched detections
for i in unmatched_dets:
trk = KalmanBoxTracker(dets[i, :])
trk.face_addtional_attribute.append(addtional_attribute_list[i])
logger.info("new Tracker: {0}".format(trk.id + 1))
self.trackers.append(trk)
i = len(self.trackers)
for trk in reversed(self.trackers):
if dets == []:
trk.update([])
d = trk.get_state()
if (trk.time_since_update < 1) and (trk.hit_streak >= self.min_hits or self.frame_count <= self.min_hits):
ret.append(np.concatenate((d, [trk.id + 1])).reshape(1, -1)) # +1 as MOT benchmark requires positive
i -= 1
# remove dead tracklet
if trk.time_since_update >= self.max_age or trk.predict_num >= predict_num or d[2] < 0 or d[3] < 0 or d[0] > img_size[1] or d[1] > img_size[0]:
if len(trk.face_addtional_attribute) >= 5:
utils.save_to_file(root_dic, trk)
logger.info('remove tracker: {0}'.format(trk.id + 1))
self.trackers.pop(i)
if len(ret) > 0:
return np.concatenate(ret)
return np.empty((0, 5))
def update(self,dets):
"""
Params:
dets - a numpy array of detections in the format [[x1,y1,x2,y2,score],[x1,y1,x2,y2,score],...]
Requires: this method must be called once for each frame even with empty detections.
Returns the a similar array, where the last column is the object ID.
NOTE: The number of objects returned may differ from the number of detections provided.
"""
self.frame_count += 1
#get predicted locations from existing trackers.
trks = np.zeros((len(self.trackers),5))
to_del = []
ret = []
colors=[]
for t,trk in enumerate(trks):
pos = self.trackers[t].predict()[0]
trk[:] = [pos[0], pos[1], pos[2], pos[3], 0]
if(np.any(np.isnan(pos))):
to_del.append(t)
trks = np.ma.compress_rows(np.ma.masked_invalid(trks))
for t in reversed(to_del):
self.trackers.pop(t)
matched, unmatched_dets, unmatched_trks = associate_detections_to_trackers(dets,trks)
#update matched trackers with assigned detections
for t,trk in enumerate(self.trackers):
if(t not in unmatched_trks):
d = matched[np.where(matched[:,1]==t)[0],0]
trk.update(dets[d,:][0])
#create and initialise new trackers for unmatched detections
for i in unmatched_dets:
trk = KalmanBoxTracker(dets[i,:])
self.trackers.append(trk)
i = len(self.trackers)
for trk in reversed(self.trackers):
d = trk.get_state()[0]
i -= 1
if((trk.time_since_update < 1) and (trk.hit_streak >= self.min_hits )):
ret.append(np.concatenate((d,[trk.id+1])).reshape(1,-1))
colors.append(trk.color)
#remove dead tracklet
elif(trk.time_since_update > self.max_age):
self.trackers.pop(i)
else:
ret.append(np.concatenate((d,[trk.id+1])).reshape(1,-1))
colors.append(trk.color)
if(len(ret)>0):
return np.concatenate(ret),colors
return np.empty((0,5)),[]
def update(self, dets, img=None):
"""
Params:
dets - a numpy array of detections in the format [[x,y,w,h,score],[x,y,w,h,score],...]
Requires: this method must be called once for each frame even with empty detections.
Returns the a similar array, where the last column is the object ID.
NOTE: The number of objects returned may differ from the number of detections provided.
"""
self.frame_count += 1
trks = np.zeros((len(self.trackers), 5))
"""
trks : array of trackers' position with the score of each tracker
to_del : array that contain indices of trackers need to be deleted as they are invalid
ret : array of returned trackers [pos,id]
colors : list of colors of the bounding boxes
"""
to_del = []
ret = []
colors = []
#get predicted locations from existing trackers.
for t, trk in enumerate(trks):
pos = self.trackers[t].predict(img)
trk[:] = [pos[0], pos[1], pos[2], pos[3], 0]
if (np.any(np.isnan(pos))):
to_del.append(t)
trks = np.ma.compress_rows(np.ma.masked_invalid(trks))
for t in reversed(to_del):
self.trackers.pop(t)
#Compare detections to trackers and fil the matched, unmatched_dets, unmatched_trks lists
matched, unmatched_dets, unmatched_trks = associate_detections_to_trackers(
dets, trks)
#update matched trackers with assigned detections
for t, trk in enumerate(self.trackers):
if (t not in unmatched_trks):
d = matched[np.where(matched[:, 1] == t)[0], 0]
trk.update(dets[d, :][0],
img) ## for dlib re-intialize the trackers ?!
#create and initialise new trackers for unmatched detections
for i in unmatched_dets:
trk = CorrelationTracker(dets[i, :], img)
self.trackers.append(trk)
i = len(self.trackers)
for trk in reversed(self.trackers):
if dets == []:
trk.update([], img)
d = trk.get_state()
if ((trk.time_since_update < 1)
and (trk.hit_streak >= self.min_hits
or self.frame_count <= self.min_hits)):
ret.append(np.concatenate((d, [trk.id + 1])).reshape(1, -1))
colors.append(trk.color)
i -= 1
#remove dead tracklet
if (trk.time_since_update > self.max_age):
self.trackers.pop(i)
else:
ret.append(np.concatenate((d, [trk.id + 1])).reshape(1, -1))
colors.append(trk.color)
if (len(ret) > 0):
return np.concatenate(ret), colors
return np.empty((0, 5)), []
def run(self, dets, img, frame_width, frame_height, timestamp=None):
"""
Params:
dets - a numpy array of detections in the format [[x,y,w,h,score],[x,y,w,h,score],...]
Requires: this method must be called once for each frame even with empty detections.
Returns the a similar array, where the last column is the object ID.
NOTE: The number of objects returned may differ from the number of detections provided.
"""
detections = np.empty((0, 6))
for obj, score, bounds in dets:
x, y, w, h = bounds
if obj.decode('utf-8') == 'person':
name = 0
else:
name = 1
detections = np.vstack(
(detections, (int(x - w / 2), int(y - h / 2), int(x + w / 2),
int(y + h / 2), score, name)))
dets = detections
self.timestamp = timestamp
self.frame_count += 1
#get predicted locations from existing trackers.
trks = np.zeros((len(self.trackers), 5))
to_del = []
ret = []
for t, trk in enumerate(trks):
pos = self.trackers[t].predict(img) #for kal!
trk[:] = [pos[0], pos[1], pos[2], pos[3], 0]
if (np.any(np.isnan(pos))):
to_del.append(t)
trks = np.ma.compress_rows(np.ma.masked_invalid(trks))
for t in reversed(to_del):
if len(self.trackers) > 0:
self.trackers.pop(t)
if dets != []:
matched, unmatched_dets, unmatched_trks = associate_detections_to_trackers(
dets, trks)
#update matched trackers with assigned detections
for t, trk in enumerate(self.trackers):
if (t not in unmatched_trks):
d = matched[np.where(matched[:, 1] == t)[0], 0]
features = fm.calculatePersonHistograms(
dets[d, :][0][0], dets[d, :][0][1], dets[d, :][0][2],
dets[d, :][0][3], img)
trk.update(dets[d, :][0], features,
img) ## for dlib re-intialize the trackers ?!
#create and initialise new trackers for unmatched detections
for i in unmatched_dets:
features = fm.calculatePersonHistograms(
dets[i][0], dets[i][1], dets[i][2], dets[i][3], img)
self.trackers.append(
KalmanBoxTracker(dets[i, :], features, dets[i, 5]))
to_be_deleted = []
for area in self.areas:
if area.closed:
area.inside = 0
for idx, trk in enumerate(reversed(self.trackers)):
if dets == []:
trk.update([], img)
d = trk.get_state()
ret.append(np.concatenate((d, [trk.id + 1])).reshape(
1, -1)) # +1 as MOT benchmark requires positive
bottom_mid_x, bottom_mid_y = self.get_bottom_mid_point(
ret[-1][-1][0], ret[-1][-1][2], ret[-1][-1][3])
trk.location_history.append(Point(bottom_mid_x, bottom_mid_y))
if trk.history_limit >= self.keep_history:
if len(self.trackers) > 0:
trk.location_history.pop(0)
else:
trk.history_limit += 1
if area.closed == True:
if len(trk.location_history) > 1:
self.change_area_count(trk, area)
else:
self.is_inside(trk, area)
#remove dead tracklet
if (trk.time_since_update > self.max_age):
pos = trk.predict(img)
features = fm.calculatePersonHistograms(
pos[0], pos[1], pos[2], pos[3], img)
closeness = fm.calculateCloseness(trk.last_features,
features)
# if distance between last time we saw this object and current estimation is more than 25. Kill it.
if closeness > 0.25:
if len(self.trackers) > 0:
to_be_deleted.append(idx)
print('Name =', area.name, 'Inside =', area.inside, 'Outside =',
area.outside, 'Counters =', area.counters)
to_be_deleted = self.unique(to_be_deleted)
self.delete_objs(to_be_deleted)
if (len(ret) > 0):
return self.areas, self.trackers
return self.areas, self.trackers
def update(self, dets, img=None):
"""
Params:
dets - a numpy array of detections in the format [[x,y,w,h,score],[x,y,w,h,score],...]
Requires: this method must be called once for each frame even with empty detections.
Returns the a similar array, where the last column is the object ID.
NOTE: The number of objects returned may differ from the number of detections provided.
"""
#get predicted locations from existing trackers.
trks = np.zeros((len(self.trackers), 5))
for t, trk in enumerate(trks):
pos = self.trackers[t].predict(img)
trk[:] = [pos[0], pos[1], pos[2], pos[3], 0]
trks = np.ma.compress_rows(np.ma.masked_invalid(trks))
if dets != []:
dets_inflated = np.copy(dets)
for i in range(dets_inflated.shape[0]):
r1 = dets_inflated[i, :]
r1 = (r1[0], r1[1], r1[2] - r1[0], r1[3] - r1[1])
r1 = model.rect_inflate2(r1, r1[2] * self.inflate_ratio,
r1[3] * self.inflate_ratio)
dets_inflated[i][0] = r1[0]
dets_inflated[i][1] = r1[1]
dets_inflated[i][2] = r1[2] + r1[0]
dets_inflated[i][3] = r1[3] + r1[1]
matched, unmatched_dets, unmatched_trks = associate_detections_to_trackers(
dets_inflated, trks)
#update matched trackers with assigned detections
for t, trk in enumerate(self.trackers):
if (t not in unmatched_trks):
d = matched[np.where(matched[:, 1] == t)[0], 0]
trk.update(dets[d, :][0], img)
#create and initialize new trackers for unmatched detections
for i in unmatched_dets:
skip = False
for tr in self.trackers:
r1 = tr.get_state()
r2 = dets[i, :]
r1 = (r1[0], r1[1], r1[2] - r1[0], r1[3] - r1[1])
r2 = (r2[0], r2[1], r2[2] - r2[0], r2[3] - r2[1])
isect = model.rect_intersection(r1, r2)
if isect:
area_factor = (isect[2] * isect[3]) / min(
r1[2] * r1[3], r2[2] * r2[3])
if area_factor > self.mix_threshold:
tr.time_since_update = 0
skip = True
break
if skip:
continue
new_id = None
for tr in self.trackers:
r2 = dets[i, :]
r2 = (r2[0], r2[1], r2[2] - r2[0], r2[3] - r2[1])
if tr.mixed_ids:
r3 = tr.get_state()
r3 = (r3[0], r3[1], r3[2] - r3[0], r3[3] - r3[1])
r3 = model.rect_inflate2(r3, max(r3[2], r3[3]),
max(r3[2], r3[3]))
isect = model.rect_intersection(r2, r3)
if isect:
new_id = tr.mixed_ids[0]
del tr.mixed_ids[0]
break
trk = KCFTracker(dets[i, :], img, new_id, self.use_4patch,
self.stop_on_4patch_break)
self.trackers.append(trk)
i = len(self.trackers)
for trk in reversed(self.trackers):
if dets == []:
trk.update([], img)
i -= 1
#remove dead tracklet
if (trk.time_since_update > self.max_age):
self.trackers.pop(i)
i = len(self.trackers)
for trk1 in reversed(self.trackers):
i -= 1
if trk1.life < self.mix_life_threshold:
continue
r1 = trk1.get_state()
r1 = (r1[0], r1[1], r1[2] - r1[0], r1[3] - r1[1])
for trk2 in self.trackers:
if trk1 == trk2:
continue
if trk2.life < self.mix_life_threshold:
continue
r2 = trk2.get_state()
r2 = (r2[0], r2[1], r2[2] - r2[0], r2[3] - r2[1])
isect = model.rect_intersection(r1, r2)
if isect:
area_factor = (isect[2] * isect[3]) / min(
r1[2] * r1[3], r2[2] * r2[3])
if area_factor > self.mix_threshold:
trk2.mixed_ids += trk1.mixed_ids
trk2.mixed_ids.append(trk1.id)
self.trackers.pop(i)
break
ret = []
for trk in reversed(self.trackers):
d = trk.get_state()
ret.append(list(d) + [trk.id, trk.points, trk.mixed_ids])
return ret
def update(self, dets, obj_classes):
"""
Params:
dets - a numpy array of detections in the format [[x,y,w,h,score],[x,y,w,h,score],...]
Requires: this method must be called once for each frame even with empty detections.
Returns the a similar array, where the last column is the object ID.
NOTE: The number of objects returned may differ from the number of detections provided.
"""
self.frame_count += 1
# get predicted locations from existing trackers.
trks = np.zeros((len(self.trackers), 5))
to_del = []
ret = []
for t, trk in enumerate(trks):
pos = self.trackers[t].predict()
# print(pos)
trk[:] = [pos[0], pos[1], pos[2], pos[3], 0]
if np.any(np.isnan(pos)):
print('isnan')
to_del.append(t)
trks = np.ma.compress_rows(np.ma.masked_invalid(trks))
for t in reversed(to_del):
self.trackers.pop(t)
print(t)
if dets != []:
matched, unmatched_dets, unmatched_trks = associate_detections_to_trackers(
dets, trks, iou_threshold=0.25)
# update matched trackers with assigned detections
for t, trk in enumerate(self.trackers):
if t not in unmatched_trks:
d = matched[np.where(matched[:, 1] == t)[0], 0]
trk.update(dets[d, :][0])
trk.tracklet_class = int(obj_classes[d])
# create and initialise new trackers for unmatched detections
for i in unmatched_dets:
trk = KalmanBoxTracker(dets[i, :])
trk.tracklet_class = int(obj_classes[i])
self.trackers.append(trk)
# update the variables used for MR tasks
for trk in reversed(self.trackers):
trk.statistics_update()
i = len(self.trackers)
for trk in reversed(self.trackers):
if dets == []:
trk.update([])
if trk.hit_streak >= self.min_hits or self.frame_count <= self.min_hits:
ret.append(trk)
i -= 1
# remove dead tracklet and counted dead tracklet
if (trk.time_since_update >
self.max_age) or (trk.count_flag and trk.time_since_counted
> self.time_since_counted_threshold):
self.trajectory_database.append(trk.trajectory)
self.trackers.pop(i)
if len(ret) > 0:
return True, ret
else:
return False, ret