forked from PaddlePaddle/PaddleDetection
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpipeline.py
1321 lines (1126 loc) · 54.8 KB
/
pipeline.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import yaml
import glob
import cv2
import numpy as np
import math
import paddle
import sys
import copy
import threading
import queue
import time
from collections import defaultdict
from datacollector import DataCollector, Result
try:
from collections.abc import Sequence
except Exception:
from collections import Sequence
# add deploy path of PaddleDetection to sys.path
parent_path = os.path.abspath(os.path.join(__file__, *(['..'] * 2)))
sys.path.insert(0, parent_path)
from cfg_utils import argsparser, print_arguments, merge_cfg
from pipe_utils import PipeTimer
from pipe_utils import get_test_images, crop_image_with_det, crop_image_with_mot, parse_mot_res, parse_mot_keypoint
from pipe_utils import PushStream
from python.infer import Detector, DetectorPicoDet
from python.keypoint_infer import KeyPointDetector
from python.keypoint_postprocess import translate_to_ori_images
from python.preprocess import decode_image, ShortSizeScale
from python.visualize import visualize_box_mask, visualize_attr, visualize_pose, visualize_action, visualize_vehicleplate, visualize_vehiclepress, visualize_lane, visualize_vehicle_retrograde
from pptracking.python.mot_sde_infer import SDE_Detector
from pptracking.python.mot.visualize import plot_tracking_dict
from pptracking.python.mot.utils import flow_statistic, update_object_info
from pphuman.attr_infer import AttrDetector
from pphuman.video_action_infer import VideoActionRecognizer
from pphuman.action_infer import SkeletonActionRecognizer, DetActionRecognizer, ClsActionRecognizer
from pphuman.action_utils import KeyPointBuff, ActionVisualHelper
from pphuman.reid import ReID
from pphuman.mtmct import mtmct_process
from ppvehicle.vehicle_plate import PlateRecognizer
from ppvehicle.vehicle_attr import VehicleAttr
from ppvehicle.vehicle_pressing import VehiclePressingRecognizer
from ppvehicle.vehicle_retrograde import VehicleRetrogradeRecognizer
from ppvehicle.lane_seg_infer import LaneSegPredictor
from download import auto_download_model
class Pipeline(object):
"""
Pipeline
Args:
args (argparse.Namespace): arguments in pipeline, which contains environment and runtime settings
cfg (dict): config of models in pipeline
"""
def __init__(self, args, cfg):
self.multi_camera = False
reid_cfg = cfg.get('REID', False)
self.enable_mtmct = reid_cfg['enable'] if reid_cfg else False
self.is_video = False
self.output_dir = args.output_dir
self.vis_result = cfg['visual']
self.input = self._parse_input(args.image_file, args.image_dir,
args.video_file, args.video_dir,
args.camera_id, args.rtsp)
if self.multi_camera:
self.predictor = []
for name in self.input:
predictor_item = PipePredictor(
args, cfg, is_video=True, multi_camera=True)
predictor_item.set_file_name(name)
self.predictor.append(predictor_item)
else:
self.predictor = PipePredictor(args, cfg, self.is_video)
if self.is_video:
self.predictor.set_file_name(self.input)
def _parse_input(self, image_file, image_dir, video_file, video_dir,
camera_id, rtsp):
# parse input as is_video and multi_camera
if image_file is not None or image_dir is not None:
input = get_test_images(image_dir, image_file)
self.is_video = False
self.multi_camera = False
elif video_file is not None:
assert os.path.exists(
video_file
) or 'rtsp' in video_file, "video_file not exists and not an rtsp site."
self.multi_camera = False
input = video_file
self.is_video = True
elif video_dir is not None:
videof = [os.path.join(video_dir, x) for x in os.listdir(video_dir)]
if len(videof) > 1:
self.multi_camera = True
videof.sort()
input = videof
else:
input = videof[0]
self.is_video = True
elif rtsp is not None:
if len(rtsp) > 1:
rtsp = [rtsp_item for rtsp_item in rtsp if 'rtsp' in rtsp_item]
self.multi_camera = True
input = rtsp
else:
self.multi_camera = False
input = rtsp[0]
self.is_video = True
elif camera_id != -1:
self.multi_camera = False
input = camera_id
self.is_video = True
else:
raise ValueError(
"Illegal Input, please set one of ['video_file', 'camera_id', 'image_file', 'image_dir']"
)
return input
def run_multithreads(self):
if self.multi_camera:
multi_res = []
threads = []
for idx, (predictor,
input) in enumerate(zip(self.predictor, self.input)):
thread = threading.Thread(
name=str(idx).zfill(3),
target=predictor.run,
args=(input, idx))
threads.append(thread)
for thread in threads:
thread.start()
for predictor, thread in zip(self.predictor, threads):
thread.join()
collector_data = predictor.get_result()
multi_res.append(collector_data)
if self.enable_mtmct:
mtmct_process(
multi_res,
self.input,
mtmct_vis=self.vis_result,
output_dir=self.output_dir)
else:
self.predictor.run(self.input)
def run(self):
if self.multi_camera:
multi_res = []
for predictor, input in zip(self.predictor, self.input):
predictor.run(input)
collector_data = predictor.get_result()
multi_res.append(collector_data)
if self.enable_mtmct:
mtmct_process(
multi_res,
self.input,
mtmct_vis=self.vis_result,
output_dir=self.output_dir)
else:
self.predictor.run(self.input)
def get_model_dir(cfg):
"""
Auto download inference model if the model_path is a url link.
Otherwise it will use the model_path directly.
"""
for key in cfg.keys():
if type(cfg[key]) == dict and \
("enable" in cfg[key].keys() and cfg[key]['enable']
or "enable" not in cfg[key].keys()):
if "model_dir" in cfg[key].keys():
model_dir = cfg[key]["model_dir"]
downloaded_model_dir = auto_download_model(model_dir)
if downloaded_model_dir:
model_dir = downloaded_model_dir
cfg[key]["model_dir"] = model_dir
print(key, " model dir: ", model_dir)
elif key == "VEHICLE_PLATE":
det_model_dir = cfg[key]["det_model_dir"]
downloaded_det_model_dir = auto_download_model(det_model_dir)
if downloaded_det_model_dir:
det_model_dir = downloaded_det_model_dir
cfg[key]["det_model_dir"] = det_model_dir
print("det_model_dir model dir: ", det_model_dir)
rec_model_dir = cfg[key]["rec_model_dir"]
downloaded_rec_model_dir = auto_download_model(rec_model_dir)
if downloaded_rec_model_dir:
rec_model_dir = downloaded_rec_model_dir
cfg[key]["rec_model_dir"] = rec_model_dir
print("rec_model_dir model dir: ", rec_model_dir)
elif key == "MOT": # for idbased and skeletonbased actions
model_dir = cfg[key]["model_dir"]
downloaded_model_dir = auto_download_model(model_dir)
if downloaded_model_dir:
model_dir = downloaded_model_dir
cfg[key]["model_dir"] = model_dir
print("mot_model_dir model_dir: ", model_dir)
class PipePredictor(object):
"""
Predictor in single camera
The pipeline for image input:
1. Detection
2. Detection -> Attribute
The pipeline for video input:
1. Tracking
2. Tracking -> Attribute
3. Tracking -> KeyPoint -> SkeletonAction Recognition
4. VideoAction Recognition
Args:
args (argparse.Namespace): arguments in pipeline, which contains environment and runtime settings
cfg (dict): config of models in pipeline
is_video (bool): whether the input is video, default as False
multi_camera (bool): whether to use multi camera in pipeline,
default as False
"""
def __init__(self, args, cfg, is_video=True, multi_camera=False):
# general module for pphuman and ppvehicle
self.with_mot = cfg.get('MOT', False)['enable'] if cfg.get(
'MOT', False) else False
self.with_human_attr = cfg.get('ATTR', False)['enable'] if cfg.get(
'ATTR', False) else False
if self.with_mot:
print('Multi-Object Tracking enabled')
if self.with_human_attr:
print('Human Attribute Recognition enabled')
# only for pphuman
self.with_skeleton_action = cfg.get(
'SKELETON_ACTION', False)['enable'] if cfg.get('SKELETON_ACTION',
False) else False
self.with_video_action = cfg.get(
'VIDEO_ACTION', False)['enable'] if cfg.get('VIDEO_ACTION',
False) else False
self.with_idbased_detaction = cfg.get(
'ID_BASED_DETACTION', False)['enable'] if cfg.get(
'ID_BASED_DETACTION', False) else False
self.with_idbased_clsaction = cfg.get(
'ID_BASED_CLSACTION', False)['enable'] if cfg.get(
'ID_BASED_CLSACTION', False) else False
self.with_mtmct = cfg.get('REID', False)['enable'] if cfg.get(
'REID', False) else False
if self.with_skeleton_action:
print('SkeletonAction Recognition enabled')
if self.with_video_action:
print('VideoAction Recognition enabled')
if self.with_idbased_detaction:
print('IDBASED Detection Action Recognition enabled')
if self.with_idbased_clsaction:
print('IDBASED Classification Action Recognition enabled')
if self.with_mtmct:
print("MTMCT enabled")
# only for ppvehicle
self.with_vehicleplate = cfg.get(
'VEHICLE_PLATE', False)['enable'] if cfg.get('VEHICLE_PLATE',
False) else False
if self.with_vehicleplate:
print('Vehicle Plate Recognition enabled')
self.with_vehicle_attr = cfg.get(
'VEHICLE_ATTR', False)['enable'] if cfg.get('VEHICLE_ATTR',
False) else False
if self.with_vehicle_attr:
print('Vehicle Attribute Recognition enabled')
self.with_vehicle_press = cfg.get(
'VEHICLE_PRESSING', False)['enable'] if cfg.get('VEHICLE_PRESSING',
False) else False
if self.with_vehicle_press:
print('Vehicle Pressing Recognition enabled')
self.with_vehicle_retrograde = cfg.get(
'VEHICLE_RETROGRADE', False)['enable'] if cfg.get(
'VEHICLE_RETROGRADE', False) else False
if self.with_vehicle_retrograde:
print('Vehicle Retrograde Recognition enabled')
self.modebase = {
"framebased": False,
"videobased": False,
"idbased": False,
"skeletonbased": False
}
self.basemode = {
"MOT": "idbased",
"ATTR": "idbased",
"VIDEO_ACTION": "videobased",
"SKELETON_ACTION": "skeletonbased",
"ID_BASED_DETACTION": "idbased",
"ID_BASED_CLSACTION": "idbased",
"REID": "idbased",
"VEHICLE_PLATE": "idbased",
"VEHICLE_ATTR": "idbased",
"VEHICLE_PRESSING": "idbased",
"VEHICLE_RETROGRADE": "idbased",
}
self.is_video = is_video
self.multi_camera = multi_camera
self.cfg = cfg
self.output_dir = args.output_dir
self.draw_center_traj = args.draw_center_traj
self.secs_interval = args.secs_interval
self.do_entrance_counting = args.do_entrance_counting
self.do_break_in_counting = args.do_break_in_counting
self.region_type = args.region_type
self.region_polygon = args.region_polygon
self.illegal_parking_time = args.illegal_parking_time
self.warmup_frame = self.cfg['warmup_frame']
self.pipeline_res = Result()
self.pipe_timer = PipeTimer()
self.file_name = None
self.collector = DataCollector()
self.pushurl = args.pushurl
# auto download inference model
get_model_dir(self.cfg)
if self.with_vehicleplate:
vehicleplate_cfg = self.cfg['VEHICLE_PLATE']
self.vehicleplate_detector = PlateRecognizer(args, vehicleplate_cfg)
basemode = self.basemode['VEHICLE_PLATE']
self.modebase[basemode] = True
if self.with_human_attr:
attr_cfg = self.cfg['ATTR']
basemode = self.basemode['ATTR']
self.modebase[basemode] = True
self.attr_predictor = AttrDetector.init_with_cfg(args, attr_cfg)
if self.with_vehicle_attr:
vehicleattr_cfg = self.cfg['VEHICLE_ATTR']
basemode = self.basemode['VEHICLE_ATTR']
self.modebase[basemode] = True
self.vehicle_attr_predictor = VehicleAttr.init_with_cfg(
args, vehicleattr_cfg)
if self.with_vehicle_press:
vehiclepress_cfg = self.cfg['VEHICLE_PRESSING']
basemode = self.basemode['VEHICLE_PRESSING']
self.modebase[basemode] = True
self.vehicle_press_predictor = VehiclePressingRecognizer(
vehiclepress_cfg)
if self.with_vehicle_press or self.with_vehicle_retrograde:
laneseg_cfg = self.cfg['LANE_SEG']
self.laneseg_predictor = LaneSegPredictor(
laneseg_cfg['lane_seg_config'], laneseg_cfg['model_dir'])
if not is_video:
det_cfg = self.cfg['DET']
model_dir = det_cfg['model_dir']
batch_size = det_cfg['batch_size']
self.det_predictor = Detector(
model_dir, args.device, args.run_mode, batch_size,
args.trt_min_shape, args.trt_max_shape, args.trt_opt_shape,
args.trt_calib_mode, args.cpu_threads, args.enable_mkldnn)
else:
if self.with_idbased_detaction:
idbased_detaction_cfg = self.cfg['ID_BASED_DETACTION']
basemode = self.basemode['ID_BASED_DETACTION']
self.modebase[basemode] = True
self.det_action_predictor = DetActionRecognizer.init_with_cfg(
args, idbased_detaction_cfg)
self.det_action_visual_helper = ActionVisualHelper(1)
if self.with_idbased_clsaction:
idbased_clsaction_cfg = self.cfg['ID_BASED_CLSACTION']
basemode = self.basemode['ID_BASED_CLSACTION']
self.modebase[basemode] = True
self.cls_action_predictor = ClsActionRecognizer.init_with_cfg(
args, idbased_clsaction_cfg)
self.cls_action_visual_helper = ActionVisualHelper(1)
if self.with_skeleton_action:
skeleton_action_cfg = self.cfg['SKELETON_ACTION']
display_frames = skeleton_action_cfg['display_frames']
self.coord_size = skeleton_action_cfg['coord_size']
basemode = self.basemode['SKELETON_ACTION']
self.modebase[basemode] = True
skeleton_action_frames = skeleton_action_cfg['max_frames']
self.skeleton_action_predictor = SkeletonActionRecognizer.init_with_cfg(
args, skeleton_action_cfg)
self.skeleton_action_visual_helper = ActionVisualHelper(
display_frames)
kpt_cfg = self.cfg['KPT']
kpt_model_dir = kpt_cfg['model_dir']
kpt_batch_size = kpt_cfg['batch_size']
self.kpt_predictor = KeyPointDetector(
kpt_model_dir,
args.device,
args.run_mode,
kpt_batch_size,
args.trt_min_shape,
args.trt_max_shape,
args.trt_opt_shape,
args.trt_calib_mode,
args.cpu_threads,
args.enable_mkldnn,
use_dark=False)
self.kpt_buff = KeyPointBuff(skeleton_action_frames)
if self.with_vehicleplate:
vehicleplate_cfg = self.cfg['VEHICLE_PLATE']
self.vehicleplate_detector = PlateRecognizer(args,
vehicleplate_cfg)
basemode = self.basemode['VEHICLE_PLATE']
self.modebase[basemode] = True
if self.with_mtmct:
reid_cfg = self.cfg['REID']
basemode = self.basemode['REID']
self.modebase[basemode] = True
self.reid_predictor = ReID.init_with_cfg(args, reid_cfg)
if self.with_vehicle_retrograde:
vehicleretrograde_cfg = self.cfg['VEHICLE_RETROGRADE']
basemode = self.basemode['VEHICLE_RETROGRADE']
self.modebase[basemode] = True
self.vehicle_retrograde_predictor = VehicleRetrogradeRecognizer(
vehicleretrograde_cfg)
if self.with_mot or self.modebase["idbased"] or self.modebase[
"skeletonbased"]:
mot_cfg = self.cfg['MOT']
model_dir = mot_cfg['model_dir']
tracker_config = mot_cfg['tracker_config']
batch_size = mot_cfg['batch_size']
skip_frame_num = mot_cfg.get('skip_frame_num', -1)
basemode = self.basemode['MOT']
self.modebase[basemode] = True
self.mot_predictor = SDE_Detector(
model_dir,
tracker_config,
args.device,
args.run_mode,
batch_size,
args.trt_min_shape,
args.trt_max_shape,
args.trt_opt_shape,
args.trt_calib_mode,
args.cpu_threads,
args.enable_mkldnn,
skip_frame_num=skip_frame_num,
draw_center_traj=self.draw_center_traj,
secs_interval=self.secs_interval,
do_entrance_counting=self.do_entrance_counting,
do_break_in_counting=self.do_break_in_counting,
region_type=self.region_type,
region_polygon=self.region_polygon)
if self.with_video_action:
video_action_cfg = self.cfg['VIDEO_ACTION']
basemode = self.basemode['VIDEO_ACTION']
self.modebase[basemode] = True
self.video_action_predictor = VideoActionRecognizer.init_with_cfg(
args, video_action_cfg)
def set_file_name(self, path):
if type(path) == int:
self.file_name = path
elif path is not None:
self.file_name = os.path.split(path)[-1]
if "." in self.file_name:
self.file_name = self.file_name.split(".")[-2]
else:
# use camera id
self.file_name = None
def get_result(self):
return self.collector.get_res()
def run(self, input, thread_idx=0):
if self.is_video:
self.predict_video(input, thread_idx=thread_idx)
else:
self.predict_image(input)
self.pipe_timer.info()
if hasattr(self, 'mot_predictor'):
self.mot_predictor.det_times.tracking_info(average=True)
def predict_image(self, input):
# det
# det -> attr
batch_loop_cnt = math.ceil(
float(len(input)) / self.det_predictor.batch_size)
self.warmup_frame = min(10, len(input) // 2) - 1
for i in range(batch_loop_cnt):
start_index = i * self.det_predictor.batch_size
end_index = min((i + 1) * self.det_predictor.batch_size, len(input))
batch_file = input[start_index:end_index]
batch_input = [decode_image(f, {})[0] for f in batch_file]
if i > self.warmup_frame:
self.pipe_timer.total_time.start()
self.pipe_timer.module_time['det'].start()
# det output format: class, score, xmin, ymin, xmax, ymax
det_res = self.det_predictor.predict_image(
batch_input, visual=False)
det_res = self.det_predictor.filter_box(det_res,
self.cfg['crop_thresh'])
if i > self.warmup_frame:
self.pipe_timer.module_time['det'].end()
self.pipe_timer.track_num += len(det_res['boxes'])
self.pipeline_res.update(det_res, 'det')
if self.with_human_attr:
crop_inputs = crop_image_with_det(batch_input, det_res)
attr_res_list = []
if i > self.warmup_frame:
self.pipe_timer.module_time['attr'].start()
for crop_input in crop_inputs:
attr_res = self.attr_predictor.predict_image(
crop_input, visual=False)
attr_res_list.extend(attr_res['output'])
if i > self.warmup_frame:
self.pipe_timer.module_time['attr'].end()
attr_res = {'output': attr_res_list}
self.pipeline_res.update(attr_res, 'attr')
if self.with_vehicle_attr:
crop_inputs = crop_image_with_det(batch_input, det_res)
vehicle_attr_res_list = []
if i > self.warmup_frame:
self.pipe_timer.module_time['vehicle_attr'].start()
for crop_input in crop_inputs:
attr_res = self.vehicle_attr_predictor.predict_image(
crop_input, visual=False)
vehicle_attr_res_list.extend(attr_res['output'])
if i > self.warmup_frame:
self.pipe_timer.module_time['vehicle_attr'].end()
attr_res = {'output': vehicle_attr_res_list}
self.pipeline_res.update(attr_res, 'vehicle_attr')
if self.with_vehicleplate:
if i > self.warmup_frame:
self.pipe_timer.module_time['vehicleplate'].start()
crop_inputs = crop_image_with_det(batch_input, det_res)
platelicenses = []
for crop_input in crop_inputs:
platelicense = self.vehicleplate_detector.get_platelicense(
crop_input)
platelicenses.extend(platelicense['plate'])
if i > self.warmup_frame:
self.pipe_timer.module_time['vehicleplate'].end()
vehicleplate_res = {'vehicleplate': platelicenses}
self.pipeline_res.update(vehicleplate_res, 'vehicleplate')
if self.with_vehicle_press:
vehicle_press_res_list = []
if i > self.warmup_frame:
self.pipe_timer.module_time['vehicle_press'].start()
lanes, direction = self.laneseg_predictor.run(batch_input)
if len(lanes) == 0:
print(" no lanes!")
continue
lanes_res = {'output': lanes, 'direction': direction}
self.pipeline_res.update(lanes_res, 'lanes')
vehicle_press_res_list = self.vehicle_press_predictor.run(
lanes, det_res)
vehiclepress_res = {'output': vehicle_press_res_list}
self.pipeline_res.update(vehiclepress_res, 'vehicle_press')
self.pipe_timer.img_num += len(batch_input)
if i > self.warmup_frame:
self.pipe_timer.total_time.end()
if self.cfg['visual']:
self.visualize_image(batch_file, batch_input, self.pipeline_res)
def capturevideo(self, capture, queue):
frame_id = 0
while (1):
if queue.full():
time.sleep(0.1)
else:
ret, frame = capture.read()
if not ret:
return
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
queue.put(frame_rgb)
def predict_video(self, video_file, thread_idx=0):
# mot
# mot -> attr
# mot -> pose -> action
capture = cv2.VideoCapture(video_file)
# Get Video info : resolution, fps, frame count
width = int(capture.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(capture.get(cv2.CAP_PROP_FRAME_HEIGHT))
fps = int(capture.get(cv2.CAP_PROP_FPS))
frame_count = int(capture.get(cv2.CAP_PROP_FRAME_COUNT))
print("video fps: %d, frame_count: %d" % (fps, frame_count))
if len(self.pushurl) > 0:
video_out_name = 'output' if self.file_name is None else self.file_name
pushurl = os.path.join(self.pushurl, video_out_name)
print("the result will push stream to url:{}".format(pushurl))
pushstream = PushStream(pushurl)
pushstream.initcmd(fps, width, height)
elif self.cfg['visual']:
video_out_name = 'output' if (
self.file_name is None or
type(self.file_name) == int) else self.file_name
if type(video_file) == str and "rtsp" in video_file:
video_out_name = video_out_name + "_t" + str(thread_idx).zfill(
2) + "_rtsp"
if not os.path.exists(self.output_dir):
os.makedirs(self.output_dir)
out_path = os.path.join(self.output_dir, video_out_name + ".mp4")
fourcc = cv2.VideoWriter_fourcc(* 'mp4v')
writer = cv2.VideoWriter(out_path, fourcc, fps, (width, height))
frame_id = 0
entrance, records, center_traj = None, None, None
if self.draw_center_traj:
center_traj = [{}]
id_set = set()
interval_id_set = set()
in_id_list = list()
out_id_list = list()
prev_center = dict()
records = list()
if self.do_entrance_counting or self.do_break_in_counting or self.illegal_parking_time != -1:
if self.region_type == 'horizontal':
entrance = [0, height / 2., width, height / 2.]
elif self.region_type == 'vertical':
entrance = [width / 2, 0., width / 2, height]
elif self.region_type == 'custom':
entrance = []
assert len(
self.region_polygon
) % 2 == 0, "region_polygon should be pairs of coords points when do break_in counting."
assert len(
self.region_polygon
) > 6, 'region_type is custom, region_polygon should be at least 3 pairs of point coords.'
for i in range(0, len(self.region_polygon), 2):
entrance.append(
[self.region_polygon[i], self.region_polygon[i + 1]])
entrance.append([width, height])
else:
raise ValueError("region_type:{} unsupported.".format(
self.region_type))
video_fps = fps
video_action_imgs = []
if self.with_video_action:
short_size = self.cfg["VIDEO_ACTION"]["short_size"]
scale = ShortSizeScale(short_size)
object_in_region_info = {
} # store info for vehicle parking in region
illegal_parking_dict = None
cars_count = 0
retrograde_traj_len = 0
framequeue = queue.Queue(10)
thread = threading.Thread(
target=self.capturevideo, args=(capture, framequeue))
thread.start()
time.sleep(1)
while (not framequeue.empty()):
if frame_id % 10 == 0:
print('Thread: {}; frame id: {}'.format(thread_idx, frame_id))
frame_rgb = framequeue.get()
if frame_id > self.warmup_frame:
self.pipe_timer.total_time.start()
if self.modebase["idbased"] or self.modebase["skeletonbased"]:
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['mot'].start()
mot_skip_frame_num = self.mot_predictor.skip_frame_num
reuse_det_result = False
if mot_skip_frame_num > 1 and frame_id > 0 and frame_id % mot_skip_frame_num > 0:
reuse_det_result = True
res = self.mot_predictor.predict_image(
[copy.deepcopy(frame_rgb)],
visual=False,
reuse_det_result=reuse_det_result,
frame_count=frame_id)
# mot output format: id, class, score, xmin, ymin, xmax, ymax
mot_res = parse_mot_res(res)
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['mot'].end()
self.pipe_timer.track_num += len(mot_res['boxes'])
if frame_id % 10 == 0:
print("Thread: {}; trackid number: {}".format(
thread_idx, len(mot_res['boxes'])))
# flow_statistic only support single class MOT
boxes, scores, ids = res[0] # batch size = 1 in MOT
mot_result = (frame_id + 1, boxes[0], scores[0],
ids[0]) # single class
statistic = flow_statistic(
mot_result,
self.secs_interval,
self.do_entrance_counting,
self.do_break_in_counting,
self.region_type,
video_fps,
entrance,
id_set,
interval_id_set,
in_id_list,
out_id_list,
prev_center,
records,
ids2names=self.mot_predictor.pred_config.labels)
records = statistic['records']
if self.illegal_parking_time != -1:
object_in_region_info, illegal_parking_dict = update_object_info(
object_in_region_info, mot_result, self.region_type,
entrance, video_fps, self.illegal_parking_time)
if len(illegal_parking_dict) != 0:
# build relationship between id and plate
for key, value in illegal_parking_dict.items():
plate = self.collector.get_carlp(key)
illegal_parking_dict[key]['plate'] = plate
# nothing detected
if len(mot_res['boxes']) == 0:
frame_id += 1
if frame_id > self.warmup_frame:
self.pipe_timer.img_num += 1
self.pipe_timer.total_time.end()
if self.cfg['visual']:
_, _, fps = self.pipe_timer.get_total_time()
im = self.visualize_video(
frame_rgb, mot_res, self.collector, frame_id, fps,
entrance, records, center_traj) # visualize
if len(self.pushurl) > 0:
pushstream.pipe.stdin.write(im.tobytes())
else:
writer.write(im)
if self.file_name is None: # use camera_id
cv2.imshow('Paddle-Pipeline', im)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
continue
self.pipeline_res.update(mot_res, 'mot')
crop_input, new_bboxes, ori_bboxes = crop_image_with_mot(
frame_rgb, mot_res)
if self.with_vehicleplate and frame_id % 10 == 0:
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['vehicleplate'].start()
plate_input, _, _ = crop_image_with_mot(
frame_rgb, mot_res, expand=False)
platelicense = self.vehicleplate_detector.get_platelicense(
plate_input)
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['vehicleplate'].end()
self.pipeline_res.update(platelicense, 'vehicleplate')
else:
self.pipeline_res.clear('vehicleplate')
if self.with_human_attr:
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['attr'].start()
attr_res = self.attr_predictor.predict_image(
crop_input, visual=False)
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['attr'].end()
self.pipeline_res.update(attr_res, 'attr')
if self.with_vehicle_attr:
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['vehicle_attr'].start()
attr_res = self.vehicle_attr_predictor.predict_image(
crop_input, visual=False)
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['vehicle_attr'].end()
self.pipeline_res.update(attr_res, 'vehicle_attr')
if self.with_vehicle_press or self.with_vehicle_retrograde:
if frame_id == 0 or cars_count == 0 or cars_count > len(
mot_res['boxes']):
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['lanes'].start()
lanes, directions = self.laneseg_predictor.run(
[copy.deepcopy(frame_rgb)])
lanes_res = {'output': lanes, 'directions': directions}
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['lanes'].end()
if frame_id == 0 or (len(lanes) > 0 and frame_id > 0):
self.pipeline_res.update(lanes_res, 'lanes')
cars_count = len(mot_res['boxes'])
if self.with_vehicle_press:
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['vehicle_press'].start()
press_lane = copy.deepcopy(self.pipeline_res.get('lanes'))
if press_lane is None:
continue
vehicle_press_res_list = self.vehicle_press_predictor.mot_run(
press_lane, mot_res['boxes'])
vehiclepress_res = {'output': vehicle_press_res_list}
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['vehicle_press'].end()
self.pipeline_res.update(vehiclepress_res, 'vehicle_press')
if self.with_idbased_detaction:
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['det_action'].start()
det_action_res = self.det_action_predictor.predict(
crop_input, mot_res)
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['det_action'].end()
self.pipeline_res.update(det_action_res, 'det_action')
if self.cfg['visual']:
self.det_action_visual_helper.update(det_action_res)
if self.with_idbased_clsaction:
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['cls_action'].start()
cls_action_res = self.cls_action_predictor.predict_with_mot(
crop_input, mot_res)
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['cls_action'].end()
self.pipeline_res.update(cls_action_res, 'cls_action')
if self.cfg['visual']:
self.cls_action_visual_helper.update(cls_action_res)
if self.with_skeleton_action:
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['kpt'].start()
kpt_pred = self.kpt_predictor.predict_image(
crop_input, visual=False)
keypoint_vector, score_vector = translate_to_ori_images(
kpt_pred, np.array(new_bboxes))
kpt_res = {}
kpt_res['keypoint'] = [
keypoint_vector.tolist(), score_vector.tolist()
] if len(keypoint_vector) > 0 else [[], []]
kpt_res['bbox'] = ori_bboxes
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['kpt'].end()
self.pipeline_res.update(kpt_res, 'kpt')
self.kpt_buff.update(kpt_res, mot_res) # collect kpt output
state = self.kpt_buff.get_state(
) # whether frame num is enough or lost tracker
skeleton_action_res = {}
if state:
if frame_id > self.warmup_frame:
self.pipe_timer.module_time[
'skeleton_action'].start()
collected_keypoint = self.kpt_buff.get_collected_keypoint(
) # reoragnize kpt output with ID
skeleton_action_input = parse_mot_keypoint(
collected_keypoint, self.coord_size)
skeleton_action_res = self.skeleton_action_predictor.predict_skeleton_with_mot(
skeleton_action_input)
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['skeleton_action'].end()
self.pipeline_res.update(skeleton_action_res,
'skeleton_action')
if self.cfg['visual']:
self.skeleton_action_visual_helper.update(
skeleton_action_res)
if self.with_mtmct and frame_id % 10 == 0:
crop_input, img_qualities, rects = self.reid_predictor.crop_image_with_mot(
frame_rgb, mot_res)
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['reid'].start()
reid_res = self.reid_predictor.predict_batch(crop_input)
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['reid'].end()
reid_res_dict = {
'features': reid_res,
"qualities": img_qualities,
"rects": rects
}
self.pipeline_res.update(reid_res_dict, 'reid')
else:
self.pipeline_res.clear('reid')
if self.with_video_action:
# get the params
frame_len = self.cfg["VIDEO_ACTION"]["frame_len"]
sample_freq = self.cfg["VIDEO_ACTION"]["sample_freq"]
if sample_freq * frame_len > frame_count: # video is too short
sample_freq = int(frame_count / frame_len)
# filter the warmup frames
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['video_action'].start()
# collect frames
if frame_id % sample_freq == 0:
# Scale image
scaled_img = scale(frame_rgb)
video_action_imgs.append(scaled_img)
# the number of collected frames is enough to predict video action
if len(video_action_imgs) == frame_len:
classes, scores = self.video_action_predictor.predict(
video_action_imgs)
if frame_id > self.warmup_frame:
self.pipe_timer.module_time['video_action'].end()
video_action_res = {"class": classes[0], "score": scores[0]}
self.pipeline_res.update(video_action_res, 'video_action')