Research On Intelligent Vehicle Path Recognition Method Based On Binocular Vision

The fast development of economy brings higher living standards to human beings in the21st century, and the demand of cars among people is becoming bigger and bigger, theautomobile has brought convenience to people, but at the same time, which also brings hugetraffic pressure to the society, causing the great increase of traffic accident rate and bringsenormous economic losses and casualties to society. It is important to study on the intelligentvehicle to avoid the accident. Then, path identification is the key to achieve intelligent vehiclewidely used. Path identification for intelligent vehicle in complex and unknown environmenthas been focused on this dissertation, the key technologies of stereo matching, lane lineidentification and extraction of structural road and unstructured road for intelligent vehiclehas been researched to improve intelligent vehicle’s intelligence, and an intelligent vehiclesystem including hardware and software has also been established.(1) In path image acquisition, binocular vision is used to achieve path identification ofintelligent vehicle. Extended kalman filtering method is used to calibrate camera for thedeformation coefficient of camera, getting the internal and external parameters of camera.(2) In path image processing, at first, carries on gray processing to the collected RGBimage, and smooth filtering to the gray image to eliminate the noise of image, Laplaciansharpening processing is employed for making the fuzzy image more clearly. Second, whichthe edge detection to the road image, Canny algorithm is used in the dissertation to extract theimage edge point, and which is improved to extract the edge of the road better by usingautomatic extraction threshold method, then, Hough transform is used to extract the lane line,and the lane line is fitted more accurately by LmedSquare, which compared with the LeastSquares and LmedSquare.(3) In vision localization, image feature(corners) detection and matching method havebeen presented. As to image feature detection and matching, SIFT algorithm is used to extractthe binocular images and match the feature point by using the certain related constraints, andthe matching accuracy is improved by using RANSAC algorithm to eliminate the falsematching, and makes3D reconstruction for the image after matching, the problems of visionlocalization and matching inaccuracy of image feature point are better solved.(4) The hardware and software of simulate intelligent vehicle have been designed andimplemented in this dissertation, based on the model car, new hardware has also beendesigned for controlling, then the intelligent vehicle platform has been built. The upper ordercomputer software is designed by VC++6.0, through the establishment of hardware platformand upper order computer software, combing the introduced path identification algorithm andtransmitting the path information handled by intelligent vehicle to the upper order computer in real time by wireless to test the accuracy of the algorithm. The experimental results showthat, the intelligent vehicle path identification system based on binocular vision achievesbetter results with a certain control strategy and path identification algorithm.