Research On Video Ranging Method Based On Multi-algorithm Fusion

With the development of science and technology and the deepening of computer vision theory,machine vision has been widely used.Binocular stereo vision ranging is used in robot navigation,vehicle navigation,industrial measurement and other fields because of its excellent ranging effect.The complete video ranging system includes image acquisition,binocular camera calibration,image preprocessing,stereo matching,target detection.Among them,binocular camera calibration,stereo matching and target detection are the research focus.In Python environment,the use of MATLAB and OpenCV to achieve a video ranging system.First of all,the calibration of binocular camera is realized by MATLAB toolbox,and the parameters of binocular camera obtained from calibration are imported into Python for processing,and the photos taken by binocular camera are pre-processed,and then combined with the binocular camera parameters,the stereo matching is realized,then the orientation of the object in the image is determined by using the target detection algorithm,and the distance between the object is obtained by combining the parallax map.The main tasks are as follows:(1)Binocular camera calibration: The principle and method of camera calibration,both linear model and non-linear model,have analyzed the advantages and disadvantages of various calibration methods.In this paper,a calibration method is used to calibrate the left camera and the right camera separately,and the traditional print calibration plate is abandoned,then,according to the parameters of the left and right cameras,the parameters of the binocular camera are derived.(2)Image pre-processing: there is a lot of noise and useless information in the images directly collected by the binocular camera,so it is slow and inaccurate to use the image directly for stereo matching,therefore,the image is preprocessed before stereo matching,and the useless information in the image is removed by grayscale firstly.The image is filtered to reduce the noise,and then Histogram equalization to enhance the image.(3)Stereo matching: stereo matching is an important part of binocular ranging,the accuracy of stereo matching determines the accuracy of the final ranging results,it includes stereo matching principle,matching primitive,matching criterion and matching algorithm.After analyzing and comparing different stereo matching algorithms,we finally choose SGBM algorithm to realize stereo matching and obtain the depth map of the image,with the calibration parameters of binocular camera,the distance in the real scene can be obtained.(4)Target detection: target detection algorithm is an important part of machine vision.This paper introduces the principle of target detection algorithm,focuses on one-stage target detection algorithm YOLO algorithm,and selects YOLOV5 algorithm,the detection of the object is realized,the position of the object in the image is obtained,and the actual distance of the object is obtained with the depth map obtained by stereo matching.