Environment Recognition Of Autonomous Navigation Vehicles Based On Binocular Stereo Vision

Autonomous navigation vehicles are robots which are capable of moving autonomously, continuously and real-timely. Research on Autonomous navigation vehicles has become a strategic target in high-tech field, which changes quickly as the demand promotion in military, civil, and scientific research areas. Environmental recognition is an important precondition for autonomous navigation vehicles making decision autonomously. As stereo vision technique has advantages of large information capacity and three-dimensional perception, it has become one of the core problems of research on autonomous navigation vehicles.In this paper, environmental recognition of autonomous navigation vehicles based on binocular stereo vision is researched, including the calibration of the binocular stereo vision system, image matching, three-dimensional reconstruction and local mapping.Firstly, the linear calibration method based on a perspective transformation model is used to calibrate the binocular stereo vision system. This method is simple, and its precision can relatively meet the request of autonomous navigation vehicles. Secondly, for the real-time demand of vehicles, the stereo matching method based on edge features and regional gradient vector features is presented. Because of the edge feature information and regional gradient vector feature information are used to match images in this method, the real-time demand of image matching and accuracy demand of environment recognition are met simultaneously. Aiming at the problem that there are too many edges in texture featured ground, the edge matching method of suspected obstacle areas based on the texture feature is proposed. Thirdly, the three-dimensional space information of the stereo matching point pairs is calculated by the three-dimensional reconstruction method based on linear calibration. Finally, three local mapping methods are presented, in which, the plane local map based on circle area is suitable for the situation of high real-time demand, and the plane local map based on polygon area and the stereo local map based on polyhedron is suitable for the situation of high accuracy demand.The feasibility of the environmental recognition method of autonomous navigation vehicles presented in this paper has been proved by simulation experiment in Matlab, and this method is worth being optimized and applied further.