POW4R: POint-Wise Full-Velocity Estimation Using 4D Radar-Camera Fusion Beyond Radial Limitations

This letter proposes an algorithm for full-velocity estimation by fusing radial velocity vectors obtained from 4D radar with optical flow vectors extracted from camera images. The full-velocity algorithm consists of a preprocessing step and a full-velocity vector estimation step. In preprocessing, radar noise is removed and ego-vehicle velocity estimation is enhanced using a Hampel filter for improved robustness in dynamic environments. In the full-velocity estimation stage, the algorithm estimates full-velocity vectors using a formulation derived from mathematical equations that incorporate multiple constraints. To evaluate the proposed method, an embedded system is implemented on a real vehicle, and datasets are collected under various scenarios. Experimental results show that the proposed algorithm significantly improves object velocity estimation performance. (error rate: baseline 81% $\rightarrow$ proposed 31%).