Opto-Electronic Engineering, Volume. 52, Issue 6, 250050(2025)
The cascade matching method for multi-object tracking of 3D LiDAR point clouds
Figures & Tables(12)
Fig. 2. Schematic diagram of AC-IoU. (a) I3D-IoU>σ; (b) I3D-IoU<σ
Fig. 4. Visualization comparison. (a) Original algorithm-scene1; (b) Proposed method-scene1; (c) Original algorithm-scene2; (d) Proposed method-scene2
Fig. 5. Visualization Comparison. (a) Original algorithm-scene3; (b) Proposed method-scene3
Fig. 6. Visualization of results. (a) Road intersection; (b) Scenarios with complex objects
Fig. 7. Visualization of results. (a) High-velocity scenes; (b) Vehicle-crowded scenes
Table 1. Tracking results of cars by different algorithms on the KITTI dataset
View tableView in ArticleTable 1. Tracking results of cars by different algorithms on the KITTI dataset
Method sAMOTA/% AMOTA/% AMOTP/% MOTA/% MOTP/% IDS FPS mmMOT 70.61 33.08 72.45 74.07 78.16 55 33 FANTrack 82.97 40.03 75.01 74.30 75.24 202 25 DeepFusionMOT 91.80 44.62 — 91.30 — 1 104 GNN3DMOT 93.68 45.27 — 84.70 — 10 5 StrongFusionMOT 93.49 46.36 — 90.68 — 0 67 EAFFMOT 93.83 47.07 77.19 86.88 76.45 15 130 FNC2 93.03 45.22 61.79 86.18 64.11 0 91 AB3DMOT 93.28 45.43 77.41 86.23 78.43 0 126 Ours 93.25 45.52 78.44 86.22 79.29 0 115
Table 2. Tracking results of pedestrians by different algorithms on the KITTI dataset
View tableView in ArticleTable 2. Tracking results of pedestrians by different algorithms on the KITTI dataset
Method sAMOTA/% AMOTA/% AMOTP/% MOTA/% MOTP/% IDS StrongFusionMOT 78.38 32.30 — 73.63 — 3 EAFFMOT 79.28 33.73 58.32 71.17 66.97 81 FNC2 61.60 21.23 35.64 56.45 50.34 28 AB3DMOT 75.85 31.04 55.53 70.90 61.27 35 Ours 81.97 33.80 60.17 73.72 64.47 2
Table 3. Tracking results of cyclists by different algorithms on the KITTI dataset
View tableView in ArticleTable 3. Tracking results of cyclists by different algorithms on the KITTI dataset
Method sAMOTA/% AMOTA/% AMOTP/% MOTA/% MOTP/% IDS StrongFusionMOT 93.39 46.36 — 90.68 — 2 EAFFMOT 93.83 47.07 77.19 86.88 76.45 15 FNC2 93.03 45.22 61.79 86.18 77.43 4 AB3DMOT 91.36 44.34 79.18 84.87 71.38 4 Ours 93.42 47.59 80.97 84.42 78.32 0
Table 4. Tracking results by different algorithms on the nuScenes dataset
View tableView in ArticleTable 4. Tracking results by different algorithms on the nuScenes dataset
Method Car Pedestrian MOTA/% MOTP/% IDS FRAG MOTA/% MOTP/% IDS FRAG StrongFusionMOT 61.30 30.34 813 533 54.30 31.61 581 165 DeepFusionMOT 57.61 32.29 981 653 48.25 35.23 591 213 AB3DMOT 63.60 31.01 515 367 62.39 24.15 376 236 Ours 66.75 32.15 356 276 64.80 31.72 295 171
Table 5. Results of ablation experiments
View tableView in ArticleTable 5. Results of ablation experiments
AC-IoU 级联匹配 sAMOTA/% AMOTA/% AMOTP/% MOTA/% MOTP/% × × 86.83 40.27 70.70 80.66 70.36 √ × 86.05 41.02 70.57 81.33 74.68 × √ 86.40 41.98 69.89 81.65 72.87 √ √ 89.55 40.30 73.19 81.45 74.03
Tools
Get Citation
Copy Citation Text
Gang Chen, Wenbiao Zhang, Hongyan Wang, En Li, Tao Sun. The cascade matching method for multi-object tracking of 3D LiDAR point clouds[J]. Opto-Electronic Engineering, 2025, 52(6): 250050
Paper Information
Category: Article
Received: Feb. 24, 2025
Accepted: Apr. 30, 2025
Published Online: Sep. 3, 2025
The Author Email: Gang Chen (陈刚)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20