Target Recognition and Range-measuring Method based on Binocular Stereo Vision

Abstract

Aiming at the problems of high cost and limited installation of traditional unmanned vehicle environment perception methods, this paper proposes a method of personnel identification and distance measurement based on the fusion of YOLOv4 and binocular stereo vision. Through the annotation of the data set, the Darknet deep learning framework is used to train and recognize the personnel, and the binocular camera disparity data is used for personnel distance detection. The experimental results show that the recognition accuracy of this method is 0.941 and the distance error is less than 5%, which can meet the task requirements of unmanned vehicle and provide technical support for solving the environment perception problems of autonomous driving vehicle.