A Method of Traffic Detection based on Millimeter-Wave Radar and Vision Fusion

An traffic detection method based on radar-vision fusion and path aggregation, named RV-PAFCOS is proposed in this paper, which extends the fully convolutional one-stage network (FCOS) framework by incorporating three key components: the radar image processing branch, the radar-vision fusion module and the path aggregation mechanism. The radar image processing branch focuses on modeling images by calibrating millimeter-wave (mmw) radar and cameras in both space and time domain, which also transforms radar point clouds into radar images for further analysis. The fusion module extracts feature from the radar and optical images based on the principle of spatial attention stitching criterion. The path aggregation module improves feature utilization by effectively integrating spatial details from shallow feature maps with high-level semantic information from deep layers. This integration mechanism enhances detection accuracy for both large-scale and small-scale targets. Through experimental analysis, the method proposed in this paper is proven on target detection performance.