Journal of Applied Optics, Volume. 45, Issue 3, 616(2024)
Low-light pedestrian detection and tracking algorithm based on autoencoder structure and improved Bytetrack
Figures & Tables(18)
Fig. 2. Model framework of multi-task autoencoding transformations (MAET)
Table 1. Multi-target tracking evaluation indicators
View tableView in ArticleTable 1. Multi-target tracking evaluation indicators
评价指标 定义 HOTA 高阶跟踪精度 MOTA 多目标跟踪准确率 IDF1 正确识别的检测与平均真实数和计算检测数之比 IDs ID切换总次数 FP 误检目标总数 FN 漏检目标总数 FPS 帧率
Table 2. Comparative experimental results of target detection algorithms
View tableView in ArticleTable 2. Comparative experimental results of target detection algorithms
算法 训练集 测试集 图像 预处理 mAP @0.5:0.95 mAP @0.5 YOLOX normal low 无 0.196 0.371 YOLOX normal low Zero DCE 0.165 0.325 YOLOX normal low Kind 0.160 0.319 YOLOX normal low URetinexNet 0.201 0.380 YOLOX low low 无 0.299 0.635 YOLOX- ASFF low low 无 0.325 0.664 Faster- RCNN low low 无 0.282 0.602 Centernet low low 无 0.284 0.599 RTMDet low low 无 0.384 0.701 本文算法 normal+low low 无 0.362 0.682
Table 3. Generalization experimental results of target detection algorithms
View tableView in ArticleTable 3. Generalization experimental results of target detection algorithms
算法 训练集 图像预 处理 mAP @0.5 mAP @0.75 Faster RCNN low 无 0.892 0.690 normal Zero DCE 0.896 0.683 normal Kind 0.860 0.683 normal URetinexNet 0.881 0.694 Centernet low 无 0.882 0.751 normal Zero DCE 0.879 0.744 normal Kind 0.864 0.732 normal URetinexNet 0.864 0.760 RTMDet low 无 0.899 0.780 normal Zero DCE 0.895 0.760 normal Kind 0.806 0.681 normal URetinexNet 0.811 0.686 YOLOX-ASFF low 无 0.945 0.760 normal Zero DCE 0.934 0.731 normal Kind 0.885 0.720 normal URetinexNet 0.905 0.746 本文算法 normal+low 无 0.949 0.795 YOLOX（基准） normal 无 0.929 0.763
Table 4. Ablation experimental results of appearance feature extraction network
View tableView in ArticleTable 4. Ablation experimental results of appearance feature extraction network
外观特征提取网络 HOTA/% MOTA/% IDF1/% osnet 70.65 88.12 85.31 transreid 72.36 89.55 88.34
Table 5. Ablation experimental results of improved components
View tableView in ArticleTable 5. Ablation experimental results of improved components
AW NSA HOTA/% MOTA/% IDF1/% — — 70.79 87.81 86.27 √ — 71.28 89.38 88.30 √ √ 72.36 89.55 88.34
Table 6. Comparison of multi-target tracking results
View tableView in ArticleTable 6. Comparison of multi-target tracking results
算法 HOTA/% MOTA/% IDF1/% IDs FP FN FPS YOLOX-Bytetrack 66.19 78.83 82.15 30 586 1412 33.93 MAET_YOLOX-Bytetrack 69.13 84.36 84.84 25 477 994 29.32 YOLOX-BytetrackAW 67.14 83.02 82.99 18 346 1249 16.37 MAET_YOLOX-BytetrackAW 72.36 89.55 88.34 15 326 651 9.98
Tools
Get Citation
Copy Citation Text
Zelin REN, Lan PANG, Chao WANG, Jiaheng LI, Fangyan ZHOU. Low-light pedestrian detection and tracking algorithm based on autoencoder structure and improved Bytetrack[J]. Journal of Applied Optics, 2024, 45(3): 616
Paper Information
Category: Research Articles
Received: Oct. 27, 2023
Accepted: --
Published Online: Jun. 2, 2024
The Author Email: ZHOU Fangyan (周方琰（1998—）)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20