Ego Vehicle Speed Estimation With Automotive Corner Radar Under Velocity Ambiguity Using Optimized MobileNetV3

Accurate estimation of ego vehicle speed is crucial for autonomous driving and advanced driver-assistance systems (ADAS). However, conventional front-facing radar-based speed estimation methods suffer from challenges such as limited field of view, Doppler ambiguity, and the presence of moving targets. Moreover, corner radar not only encounters these issues but also faces additional difficulties due to its mount angle, leading to severe performance degradation in complex environments. To overcome these limitations, this study proposes a lightweight deep learning-based regression model optimized for corner radar, enabling robust and real-time speed estimation. The proposed method utilizes the MobileNetV3-Small architecture to achieve high accuracy while maintaining computational efficiency. Experimental validation using real-world measurement data demonstrates that the proposed model surpasses conventional methods, exhibiting strong generalization capability even in environments with velocity ambiguity. These findings confirm the feasibility of radar-only speed estimation without additional vehicle sensors and emphasize the potential of corner radar beyond object detection, extending its application to autonomous driving systems.