Radar-Vision Fusion for Vehicle Detection and Tracking

The fusion of multi-sensors can prevent the drawbacks of a single sensor. In this paper, we propose a method to fuse tracked radar and vision targets. We use DBSCAN and SORT algorithms to detect and track radar targets and apply YOLOv5 and DeepSORT algorithms to generate vision target boxes. Then, we execute the time and spatial fusion for radar and vision targets. We conduct experiments on the Oxford Radar RobotCar Dataset and prove the effectiveness of our method.