Research Of Obstacle Detection Preceding Vehicle Based On Binocular Vision

Reliable obstacle recognition and location technology can help drivers understand the road information in advance.When driving in the lower visible road conditions,due to the lack of timely evasion,it is very likely to lead to traffic accidents.In recent years,obstacle detection based on machine vision has become a hot spot,binocular stereo vision technology has been used by most scholars as an important auxiliary means of intelligent driving research because of its irreplaceable advantages.For example,it is used to detect obstacles in front of vehicles.This paper first analyzes the research status of binocular stereo vision and obstacle detection technology both at home and abroad,and chooses binocular vision technology to detect obstacles;then an improved stereo matching algorithm based on multi-scale cost model framework(Cross-Scale Cost Aggregation,CSCA)is proposed.Then using ASW algorithm,SGBM algorithm,GF+WM(guiding filtering combined with weighted median filtering algorithm)and the proposed algorithm of this paper to stereo matching using Middlebury platform data sets,calculating and comparing the mismatch rate and running time.A distance test in real environment is designed.This paper mainly studies the key technologies in the realization of binocular vision,and details about how to achieve the obstacle detection in front of the vehicle are as follows:1.Build binocular vision platform.By analyzing the parallel binocular stereo vision model and the intersecting binocular stereo vision model,combined with the shooting vision and search area conditions,we choose the parallel binocular stereo vision model as the visual model of the system.Integrated with all kinds of hardware and software equipment,a double stereo vision obstacle detection platform is built.2.Binocular camera calibration and correction.The transformation relationship between the four kinds of coordinate systems is combed,and the mathematical model of the transformation from the world coordinate system to the image coordinate system is calculated.According to the conditions of camera,choose to use the Zhang Zhengyou calibration method for camera calibration that can get the internal and external parameters of the cameras;On this basis,the relationship between calibration distance and calibration accuracy is studied,and the optimal calibration distance is selected to minimize the calibration error.The Bouguet method is used to revise the camera.3.Stereo matching for images.Study the four steps of stereo matching algorithm and various methods of cost aggregation are analyzed and compared.Introduce the cross-scale cost aggregation model and propose an improved algorithm based on this framework.Compared the reliability of the algorithm with ASW algorithm,SGBM algorithm and GF + WM algorithm by mismatching rate and running time.4.Acquisition of image depth information.VS2012 and Opencv2.4.9 software are used as the development platform,and the algorithm in this paper combining the camera calibration parameters is used to obtain the depth information of the image.In order to further verify the reliability of the matching algorithm,the distance test of different distance vehicles in front of the camera is designed.The range of distance is 1 to 10 meters.Use the proposed algorithm,SGBM algorithm and GF+WM algorithm to analyze the actual distance of the vehicle setting at different distance and analyze the result of the test.The experimental results show that the distance error rate is reduced as a whole compared with the other two algorithms,and the error of the range is larger with the increase of the distance between the two cameras.When the distance between objects is within 3 to 6 meters,the error rate is relatively low.It is reliable to detect obstacles in this range.