The KITTI Vision Benchmark Suite

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.

\

@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},
}