Zhuohao Yan

Publications

LIO-LOT teaser: LiDAR tracking on KITTI
LIO-LOT: Tightly-Coupled Multi-Object Tracking and LiDAR-Inertial Odometry
X. Li, Z. Yan, S. Feng†, et al.
IEEE Transactions on Intelligent Transportation Systems, 2024. Student first author.

Puts MOT and LiDAR-inertial odometry in one factor graph. Objects are associated with a hybrid matching strategy, then split into high- and low-confidence tracks and hierarchically optimized with IMU and static structure. Stronger ego-localization and tracking on KITTI and nuScenes, especially under occlusion and truncation.

IEEE
S3MOT KITTI demo
S3MOT: Monocular 3D Object Tracking with Selective State Space Model
Z. Yan, S. Feng, X. Li†, et al.
arXiv:2504.18068, 2025.

A monocular 3D tracker built on selective state-space models. HSSM, VeloSSM, and FCOE fuse appearance, motion, and spatial cues into a differentiable association model, reaching 76.86 HOTA at 31 FPS on the KITTI test set.

arXiv Code
LiDAR-vision MOT: multimodal feature matching
Tightly Coupled Integration of LiDAR and Vision for 3D Multi-object Tracking
S. Feng, X. Li†, Z. Yan, et al.
IEEE Transactions on Intelligent Vehicles, 2024. Student second author.

Tightly fuses LiDAR and camera detections for 3D multi-object tracking, so each sensor covers the other's failure modes. Joint association and state estimation improve robustness when one modality is noisy or incomplete.

IEEE
FGO-MOT KITTI demo
Accurate and Real-Time 3D-LiDAR Multi-Object Tracking Using Factor Graph Optimization
S. Feng, X. Li†, Z. Yan, et al.
IEEE Sensors Journal, 24(2):1760–1771, 2023. Student second author.

A real-time 3D LiDAR MOT system that casts data association and object-state estimation as factor graph optimization, refining tracks jointly rather than in a separate filter.

IEEE