A Traffic Scene Object Detection Method Combining Deep Learning and Stereo Vision Algorithm

Abstract

Object detection has been an important topic in the field of intelligent vehicle. The deep learning object detection is based on monocular vision without depth of the scene information and the detection of small objects in complex traffic scenes often has problems such as misdetection and omission, so the detection of small objects in traffic scenes is still a big challenge. In this paper, a method combining depth information with YOLOv5s object detection algorithm is proposed to improve the accuracy of object detection. Firstly, the depth information is obtained by the disparity images which generated by the end-to-end PSMNet network. Secondly, add an Attention Feature Fusion Module (AFFM) to YOLOv5s to improve the accuracy of small object detection. Finally, the depth information is fused with the object detection to reduce the probability of missed detection and false detection, and the distance information is also obtained. The experimental results show that the SUPER_YOLOv5s object detection algorithm combined with stereo vision can reduce the rate of missed detection and improve the detection accuracy.