Research2022 – 2025

RaLiBEV — Radar & LiDAR BEV Fusion

Co-first author · IEEE TCSVT 2025

Architecture: LiDAR and radar range–azimuth features meet in an interactive transformer-based BEV fusion module, then an anchor-free head predicts oriented boxes. IEEE TCSVT paper

Schematic

Radar penetrates rain and fog where LiDAR degrades. RaLiBEV fuses radar range–azimuth heatmaps with LiDAR point clouds in a bird's-eye-view representation, using an interactive transformer-based fusion module and a consistency-aware label assignment strategy for anchor-box-free detection.

It outperforms prior methods by 13.1% in clear weather and 19.0% in fog at IoU 0.8.

Highlights

Interactive transformer fusion of radar and LiDAR
Consistency-aware label assignment, anchor-box-free
+19.0% in fog at IoU 0.8

IEEE Xplore arXiv Code

Publication

2025Journal

★RaLiBEV: Radar and LiDAR BEV Fusion Learning for Anchor-Box-Free Object Detection Systems

Y. Yang*, J. Liu*, T. Huang, Q.-L. Han, G. Ma, B. Zhu

IEEE Transactions on Circuits and Systems for Video Technology · 35(5):4130–4143 · *equal contribution

Fuses radar range–azimuth heatmaps with LiDAR in bird's-eye view for all-weather detection: +13.1% (clear) and +19.0% (fog) at IoU 0.8.

IEEE arXiv Code Project