Free Space Detection Using Automotive Radar

This is the authors' abstract. We don't add key points for this paper.

This paper proposes a novel free space detection method based on Occupancy Grid Map (OGM) with automotive radars. Because radar sensors on commercial vehicles provide sparse detections, the limitations of radar sensors become clearly apparent in existing OGM generation methods. For generating high-quality OGM with automotive radars, we augment additional radar detections using radar attributes such as Doppler velocity and bounding boxes and use them as inputs of a deep learning-based OGM generation model. We also apply a Bayesian framework to enable OGM generation that considers the dynamic changes in the driving environment. The performance of the proposed method is verified through experiments using real driving data, demonstrating significantly improved performance compared to the result with sparse radar detections.