Instantaneous Longitudinal Ego-Velocity Estimation Using 4-D Radar–Camera Fusion in Dynamic Road Environments

Accurate ego-velocity estimation is crucial for autonomous driving systems as it supports key tasks such as odometry and simultaneous localization and mapping. In this paper, we propose a novel framework for longitudinal ego-velocity estimation from a single frame using 4D imaging radar-camera fusion. The proposed method mitigates the influence of dynamic objects on ego-velocity estimation by integrating radar and camera data. Our framework associates objects detected from radar and camera data and removes dynamic objects identified by the camera. Longitudinal ego-velocity is estimated within a single frame using weighted histogram analysis and is further refined through cosine curve fitting. Experimental results in real driving environments show that the proposed method reduces estimation errors and failure rates compared to conventional methods, while achieving accurate ego-velocity estimation from single-frame data.