Method

Rethink Multiple Object Tracking [Rethink MOT]


Submitted on 29 May. 2022 14:40 by
leichen wang (Robert Bosch CN)

Running time:0.3 s
Environment:4 cores @ 2.5 Ghz (Python)

Method Description:
In this work we rethink the widely used“tracking
by detection"(TBD) paradigm. We analyze several
typical tracking failure cases of existing TBD
algorithms.

To solve those failure cases, we design three
novel modules: 1) confidence-guided tracking
before detection module, 2) trajectory outlier
detection module, 3) motion-model-based trajectory
smoother.

Comprehensive experimental results on KITTI
Dataset and nuScenes Dataset shows that our final
method could achieve new SOTA results with these
plug-in modules.
Parameters:
alpha=0.2
Latex Bibtex:
@inproceedings{wang2023mot,
title={Towards Robust Reference System for
Autonomous Driving: Rethinking 3D MOT},
author={Wang, Leichen and Zhang, Jiadi and Cai, Pei
and Li, Xinrun},
booktitle={Proceedings of the 2023 IEEE
International Conference on Robotics and Automation
(ICRA)},
year={2023}
}

Detailed Results

From all 29 test sequences, our benchmark computes the commonly used tracking metrics CLEARMOT, MT/PT/ML, identity switches, and fragmentations [1,2]. The tables below show all of these metrics.


Benchmark MOTA MOTP MODA MODP
CAR 91.47 % 85.63 % 91.68 % 88.28 %

Benchmark recall precision F1 TP FP FN FAR #objects #trajectories
CAR 96.63 % 95.95 % 96.29 % 37136 1567 1295 14.09 % 46680 791

Benchmark MT PT ML IDS FRAG
CAR 89.38 % 6.31 % 4.31 % 72 180

This table as LaTeX


[1] K. Bernardin, R. Stiefelhagen: Evaluating Multiple Object Tracking Performance: The CLEAR MOT Metrics. JIVP 2008.
[2] Y. Li, C. Huang, R. Nevatia: Learning to associate: HybridBoosted multi-target tracker for crowded scene. CVPR 2009.


eXTReMe Tracker