A Novel Multi-Modal Radar-Vision Fusion for Robust Object Detection in Autonomous Driving

This paper presents a comprehensive study on object detection methods utilizing the fusion of millimeter-wave (mmWave) radar and visual sensors. The research addresses the inherent limitations of single sensors, such as detection accuracy and environmental robustness, as well as the disparities in data dimensionality, output formats, and operating frequencies between different sensor types. We propose and investigate multi-modal feature fusion techniques for road object detection using both mmWave radar point clouds and raw 4D radar data, integrating with visual information. The proposed fusion detection models significantly enhance overall detection accuracy and environmental robustness, meeting the application requirements for challenging scenarios including adverse weather conditions and single-sensor failure.