Obstacle Distance Detection System Based On Binocular Vision

With the development of automatic navigation technologies such as robots and unmanned aerial vehicles,the requirements for detection of the surrounding environment in the automatic navigation process have also increased,especially the accurate and real-time detection of obstacle distance information.Measuring obstacles through binocular vision technology is one of the hot research fields,and this technology is a non-contact ranging method.The main process : use the binocular camera as the image number acquisition unit to shoot the actual scene,then use the relevant algorithm to match the collected left and right images to obtain the disparity map,and finally use the disparity map to calculate the distance and other information.In order to achieve fast,accurate and convenient measurement of obstacle distance information,based on binocular vision technology,this thesis designs a distance measurement system based on an embedded system,which provides a technical basis for the automatic navigation system to avoid obstacles.Firstly,this thesis studies the imaging model of binocular vision system,and explains how to use the parallax value to calculate the distance of binocular system through the ideal imaging model.Due to the distortion of the camera itself,the actual imaging model is not ideal,which leads to the camera distortion model.In order to obtain the parameters of the binocular camera,it is necessary to calibrate the camera.After comparing various calibration methods,this thesis uses Zhang Zhengyou calibration method to calibrate the binocular camera to obtain the relevant parameters,and analyzes the errors in the process of camera calibration experiment.In order to reduce the influence caused by camera distortion,the image data is corrected and good results are achieved.Then focus on the research of the census transform matching algorithm,Aiming at the low matching accuracy of the traditional Census transform matching algorithm in weak texture regions and the susceptibility to noise points,as well as improving the real-time performance of the Census algorithm,an improved Census transform and adaptive window stereo matching algorithm is proposed.Three states of census transform are used to improve the anti-interference performance.According to the fluctuation of regional texture,the algorithm window size is adjusted to improve the accuracy and real-time performance.The performance of the algorithm is tested on the standard image of Middlebury platform.The experimental results show that the average error matching rate reaches 5.51%,which improves the problem of low accuracy to a certain extent.Compared with the traditional census algorithm,the average time spent is reduced by 36.60%,which improves the real-time performance of the algorithm.Finally,a flexible ranging system based on embedded platform is designed to solve the problem that the current binocular ranging technology is limited by the computer platform.The system is a kind of intelligent car which is composed of raspberry pie STM32.The integrated binocular camera is selected as the data acquisition module.The improved census transform matching algorithm is transplanted to raspberry pie to calculate the distance information of obstacles,and the calculation results are transmitted to the single chip microcomputer through the serial to control the intelligent car,so as to achieve convenient and fast measurement of distance information of obstacles.