Preceding Vehicle Detection Method Based on Information Fusion of Millimetre Wave Radar and Deep Learning Vision

Abstract

In the preceding vehicle detection based on the fusion of millimetre wave radar and camera, in order to obtain more accurate front vehicle status, a preceding vehicle detection method based on the information fusion of millimetre wave radar and deep learning vision is proposed in this paper. In this method, millimetre wave radar and camera are used to collect the information of preceding vehicle target. Firstly, the space-time coordinates of millimetre wave radar and camera are unified by imaging principle, coordinate conversion and same frame data selection. Then millimetre wave radar senses the state of preceding vehicle and divides the region of interest (ROI) by using the aspect ratio of vehicle. The camera uses YOLOv3-tiny to the ROI, and the deep learning algorithm of neural network model achieves high precision target detection and classification. Finally, the information fusion under different conditions is carried out according to the vehicle detection results of the two sensors and their Intersection- over-Union (IoU). The experimental results show that the recognition rate of the fusion method for typical vehicles is over 90%, and more accurate vehicle information can be obtained.