Chinese Journal of Liquid Crystals and Displays, Volume. 38, Issue 8, 1128(2023)
Stable and real-time pedestrian tracking method based on improved DeepSORT under complex background
Fig. 1. Pedestrian multi-objective algorithm framework proposed in this paper
Fig. 9. Visualization of tracking effect of improved algorithm in MOT16-02
Table 1. Evaluation index of multi-target tracking
View tableView in ArticleTable 1. Evaluation index of multi-target tracking
指标 含义 MOTA 跟踪精度，衡量目标轨迹的保持程度 HOTA 更高维跟踪精度，更加重视目标检测和
数据关联精度的平均性测量
IDF1 衡量跟踪器身份维持能力 MT 整个视频中超过80%的时间被正确
跟踪的轨迹个数
ML 整个视频中不超过20%时间被正确
跟踪的轨迹个数
IDSW 目标身份的切换次数 Frages 轨迹碎片的个数
Table 2. Ablation experiment results of three models on MOT16 dataset
View tableView in ArticleTable 2. Ablation experiment results of three models on MOT16 dataset
模型 HOTA↑/% MOTA↑/% IDF1↑/% IDSW↓ MT↑/% ML↓/% Frages↓ YOLOv5+DeepSORT 43.3 53.5 51.0 1 051 29.6 26.5 1 173 YOLOv5+CoorATT+DeepSORT 51.2 60.6 57.1 906 32.9 24.4 948 YOLOv5+CoorATT+DeepSORT+KCF 57.1 66.5 64.2 641 35.0 21.7 620
Table 3. Comparison results between the algorithm in this paper and other advanced algorithms on MOT16 dataset
View tableView in ArticleTable 3. Comparison results between the algorithm in this paper and other advanced algorithms on MOT16 dataset
HOTA↑/% MOTA↑/% IDF1↑/% IDSW↓ MT↑/% ML↓/% FPS↑ Tracktor++［ 13 ］47.3 56.3 55.1 1 247 19.5 36.6 15 CenterTrack［ 20 ］48.8 60.7 57.2 1 309 32.9 23.1 13.8 DeepMOT［ 21 ］42.4 53.7 53.8 1 647 19.1 37.0 4.9 TubeTK［ 22 ］48.7 63.2 58.6 1 137 33.5 19.4 1.0 FairMOT［ 23 ］55.3 65.3 66.7 874 34.7 20.9 23.7 本文方法 57.1 66.5 64.2 641 35.0 21.7 22.6
Table 4. Comparison results between the algorithm in this paper and other advanced algorithms on MOT17 dataset
View tableView in ArticleTable 4. Comparison results between the algorithm in this paper and other advanced algorithms on MOT17 dataset
HOTA↑/% MOTA↑/% IDF1↑/% IDSW↓ MT↑/% ML↓/% FPS↑ Tracktor++［ 13 ］51.9 53.5 52.3 2 706 17.9 35.2 14 CenterTrack［ 20 ］50.2 61.5 57.4 2 898 27.4 24.8 17 DeepMOT［ 18 ］46.4 54.2 55.7 3 947 18.6 37.9 6.9 TubeTK［ 22 ］48.0 62.0 58.6 4 137 29.1 22.9 3.0 FairMOT［ 23 ］57.4 67.3 62.5 3 303 33.2 20.3 22.9 本文方法 58.3 65.7 62.8 2 072 33.8 19.9 24
Get Citation
Copy Citation Text
Li-juan ZHANG, Zi-wei ZHANG, Yu-tong JIANG, Dong-ming LI, Meng-da HU, Ying-xue LIU. Stable and real-time pedestrian tracking method based on improved DeepSORT under complex background[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(8): 1128
Category: Research Articles
Received: Oct. 22, 2022
Accepted: --
Published Online: Oct. 9, 2023
The Author Email: Yu-tong JIANG (jiangyutong201@163.com)
