Laser & Optoelectronics Progress, Volume. 61, Issue 18, 1837003(2024)
Hierarchical Matching Multi-Object Tracking Algorithm Based on Pseudo-Depth Information
Figures & Tables(10)
Fig. 5. Experimental results of occlusion coefficient. (a) HOTA experimental results on MOT17 validation set; (b) AssA experimental results on MOT17 validation set; (c) IDF1 experimental results on MOT17 validation set; (d) HOTA experimental results on DanceTrack test set; (e) AssA experimental results on DanceTrack test set; (f) IDF1 experimental results on DanceTrack test set
Fig. 6. Tracking results of the DanceTrack data sets. (a) DanceTrack0067; (b) DanceTrack0070
Table 1. Evaluation indicators and instructions
View tableTable 1. Evaluation indicators and instructions
Evaluation indicator Indicator instruction HOTA［ 15 ］Unified metric for multi-object tracking，balancing detection，association，and localization accuracy. DetA Measures object detection accuracy. AssA Assesses object association accuracy MOTA Tracking accuracy. Calculated by integrating FP，FN，IDS，and other indicators IDF1 Ratio of correctly identified detections to the average number of true and calculated detections FPS Number of images processed per second
Table 2. Experimental results on the MOT17 test set
View tableTable 2. Experimental results on the MOT17 test set
Model HOTA /% DetA /% AssA /% MOTA /% IDF1 /% FPS /（frame/s） FairMOT 59.3 60.9 58.0 73.7 72.3 13.2 QDTrack［ 16 ］53.9 55.6 52.7 68.7 66.3 8.5 TransTrack［ 17 ］54.1 61.6 47.9 75.2 63.5 7.2 TraDes［ 18 ］52.7 55.2 50.8 69.1 63.9 7.0 MOTR 57.2 58.9 55.8 71.9 68.4 <7.5 GTR［ 19 ］59.1 61.6 57.0 75.3 71.5 11.2 ByteTrack 63.1 64.5 62.0 80.3 77.3 17.5 Proposed 65.1 65.3 65.2 81.0 80.1 12.3
Table 3. Experimental results on the DanceTrack test set
View tableTable 3. Experimental results on the DanceTrack test set
Model HOTA /% DetA /% AssA /% MOTA /% IDF1 /% FPS /（frame/s） FairMOT 39.7 66.7 23.8 82.2 40.8 13.2 QDTrack 54.2 80.1 36.8 87.7 50.4 8.5 TransTrack 45.5 75.9 27.5 88.4 45.2 7.2 TraDes 43.3 74.5 25.4 86.2 41.2 7.0 MOTR 54.2 73.5 40.2 79.9 51.5 <7.5 GTR 48.0 72.5 31.9 84.7 50.3 11.2 SparseTrack 55.5 78.9 39.1 89.6 58.3 12.5 ByteTrack 47.7 71.0 32.1 89.6 53.9 17.5 Proposed 58.5 79.0 43.4 89.6 59.8 12.3
Table 4. Results of the ablation experiment on ByteTrack model
View tableTable 4. Results of the ablation experiment on ByteTrack model
Module MOT17 validation set DanceTrack test set HOTA DetA AssA IDF1 HOTA DetA AssA IDF1 Original 67.9 66.6 69.8 79.4 47.5 71.1 31.9 53.6 HDM（IOU） 68.8 66.8 71.4 81.0 56.0 78.3 40.1 58.8 IOU-D 68.5 66.6 70.5 80.0 53.1 71.9 39.4 59.5 HDM（IOU）+IOU-D 69.4 66.9 72.6 81.9 58.5 79.0 43.4 59.8
Tools
Get Citation
Copy Citation Text
Peng Hu, Shuguo Pan, Wang Gao, Ping Wang, Peng Guo. Hierarchical Matching Multi-Object Tracking Algorithm Based on Pseudo-Depth Information[J]. Laser & Optoelectronics Progress, 2024, 61(18): 1837003
Paper Information
Category: Digital Image Processing
Received: Dec. 22, 2023
Accepted: Jan. 26, 2024
Published Online: Sep. 9, 2024
The Author Email: Shuguo Pan (psg@seu.edu.cn)
CSTR:32186.14.LOP232725
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20