Method

Motion-Compensated Multi-Sensor Fusion with Hierarchical Association for 3D Multi-Target Tracking [MCMSF-HA-3DMTT]


Submitted on 6 Aug. 2024 10:13 by
(University of South China)

Running time:0.01 s
Environment:1 core @ 2.5 Ghz (C/C++)

Method Description:
We present MCMSF-HA-3DMTT, a novel framework that
achieves robust 3D multi-target tracking through
motion-compensated sensor fusion and hierarchical
association. First, high-precision spatial and
feature alignment is performed between LiDAR point
clouds and a forward-facing monocular camera, and
a dynamic confidence-weighted fusion strategy is
applied to enhance the robustness of the 3D
detector. Next, a coupled-state Kalman filter
recursively estimates target states while
compensating for ego-vehicle motion to eliminate
platform-induced interference. For data
association, we introduce a multi-level spatial
indexing structure; at each level, geometric and
appearance cues are jointly fused, and targets are
matched via a stage-wise gating and assignment
protocol. Extensive experiments on public
benchmarks demonstrate that MCMSF-HA-3DMTT
significantly improves tracking accuracy and
stability in complex driving scenarios.
Parameters:
\
Latex Bibtex:
@inproceedings{zhang2026motion,
title = {Motion-Compensated Multi-Sensor
Fusion with Hierarchical Association for 3D Multi-
Target Tracking},
author = {Zhang, Mengyao and Jiang, Chao and
Zhang, Mingyue},
booktitle = {Proceedings of the AAAI Conference
on Artificial Intelligence (AAAI)},
year = {2026},
note = {Submitted for review; under
consideration},
}

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 80.83 % 78.73 % 80.87 % 83.71 %
PEDESTRIAN 58.48 % 71.14 % 60.47 % 90.74 %

Benchmark recall precision F1 TP FP FN FAR #objects #trajectories
CAR 90.87 % 91.71 % 91.29 % 34470 3116 3462 28.01 % 42685 1651
PEDESTRIAN 77.76 % 82.32 % 79.97 % 18271 3925 5226 35.28 % 27439 1002

Benchmark MT PT ML IDS FRAG
CAR 73.85 % 22.92 % 3.23 % 16 330
PEDESTRIAN 53.26 % 36.77 % 9.97 % 460 1323

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