Chinese Journal of Liquid Crystals and Displays, Volume. 37, Issue 7, 880(2022)
Pedestrian multi-target tracking method based on YOLOv5 and person re-identification
Figures & Tables(14)
Fig. 5. Contrast of the target ID before and after frame in the tracking process using the prealgorithm and improved algorithm
Table 1. Feature extraction network experiments（Top-1）
View tableView in ArticleTable 1. Feature extraction network experiments（Top-1）
Model Market1501 GAN（Market1501） ResNeXt50 98.2% 98.8%
Table 2. Ablation of YOLOv5s
View tableView in ArticleTable 2. Ablation of YOLOv5s
Model Precision/% Recall/% mAP/% mAP@0.5∶0.95/% YOLOv5 93.5 92.1 95.8 76.4 YOLOv5+上采样 98.6 95.7 99.4 87.1
Table 3. Comparative results on this grade algorithm and other advanced algorithm on MOT16 data set
View tableView in ArticleTable 3. Comparative results on this grade algorithm and other advanced algorithm on MOT16 data set
Algorithm MOTA MOTP IDF1 MT/% ML/% IDs↓ 帧率/Hz SORT［ 3 ］59.8 79.6 53.8 25.4 22.7 1 423 8.6 DeepSORT［ 4 ］61.4 79.1 62.2 32.8 18.2 781 6.4 JDE［ 6 ］64.4 — 55.8 35.4 20.0 1 544 18.5 POI［ 23 ］66.1 79.5 65.1 34.0 20.8 805 5.0 TubeTK［ 24 ］64.9 59.4 59.4 33.5 19.4 1 117 1.0 Ours 66.2 80.8 65.8 35.3 17.6 760 5.8
Tools
Get Citation
Copy Citation Text
Yu-ting HE, Jin CHE, Jin-man WU. Pedestrian multi-target tracking method based on YOLOv5 and person re-identification[J]. Chinese Journal of Liquid Crystals and Displays, 2022, 37(7): 880
Paper Information
Category:
Received: Jan. 24, 2022
Accepted: --
Published Online: Jul. 7, 2022
The Author Email: Jin CHE (koalache@126.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20