Comparative Analysis of mmWave Radar-based Object Detection in Autonomous Vehicles

Ziaaddin Sharifisoraki, Marzieh Amini, Sreeraman Rajan

Abstract: Millimeter-wave radar technology is gaining popularity as a perception sensor in autonomous vehicles. This is due to its ability to detect nearby objects in adverse weather conditions, such as rain, snow, or fog, as well as its cost-effectiveness. In this paper, we explore the impact of different backbones and object detector heads on the performance of radar-based object detection algorithms. More specifically, we employ the RADDet dataset and its object detection algorithm which provides the entire Range-Azimuth-Doppler spectrum and incorporates an automatic annotation approach. We examine different backbones and object detector heads to identify optimal model combinations for autonomous driving applications. Our results show that using a YOLOv4 head integrated with a modified ResNet backbone leads to the highest mean average precision, reaching 66.3% with an intersection over union (IoU) of 0.1, and 43.6% with an IoU of 0.3. This observation will help to advance radar-based object detection, thereby enhancing safety and reliability in diverse environmental conditions.