Research On The Method Of Three-dimensional Information Restoration Based On Binocular Stereo Vision

With the development of society and the rapid development of the industry,machine vision technology has continuously been applying to all fields in life.Machine vision technology is gradually playing an important role in social economicdevelopment. One of the important branches in machine vision is stereo vision whichis an important symbol of intelligence. In the machine, vision system is as importantas the eyes for a man. Therefore, in this paper, the stereo binocular vision system is tobe the research object and the3-d information recovery is to be the ultimate goal. Themain process is recovering2-d image information getting by cameras to3-dinformation through a series of processing and operation. Letting the machine bodyperceive external depth information liking the geometry and posture of objects and therelative position between objects and so on in3-d environment, so it can make thedecision needed to objective objects.The basic principle of binocular stereo vision technology research is imitatingthe three-dimensional perception process of human vision. By acquiring images fromdifferent viewing angles in the same scene using two cameras and using triangulatemeasuring principle to calculate the parallax between two corresponding matchingpoints in the images, thereby the depth information of objects is acquired in the imagescene.In this paper the binocular stereo vision research contents include five partswhich are image acquisition, camera calibration, image processing, three-dimensionalmatching and the depth recovery. One camera calibration and image preprocessing arethe main research objects in this paper.Firstly, the image is acquired by two CCD cameras equipped with the binocularvision device.it needs to calibrate the camera system because of the error of systemitself. According to the actual situation of this paper, Zhengyou Zhang calibrationalgorithm is used for camera calibration.The method is simple and easy to operate,and has high precision and no strict requirement for calibration. It only needs a standard2D checkerboard plane.It is a calibration between traditional calibration and selfcalibration, which separates internal and external parameters. First, it uses videocameras to get the coordinates of the feature point on the checkerboard template andthe camera imaging model to calibrate the internal parameter matrix and the initialvalue of external parameter. Second, it is to calibrate the initial value of distortionparameters and optimize the inside and outside parameters of all cameras.Secondly, it is to deal with the obtained images, which are usually containingamount of information and also mixing with different noise likely. Therefore, beforethe stereo matching, the complicated pictures need to be simplified, which facilitatethe machine body to store, calculate and recognize. Mainly including several parts:which are to gray color images to reduce the data of images for easy storage, tosmooth and remove noise to reduce the noise interference to image, to raise the imagecontrast by image augmented to outstand the definition of goal body for facilitatingmachine recognition, to extract the edge of the image by edge detection to bepreparation for the next stereo matching.Finally,it is to be stereo matching and depth information recovery. According tothe principle of epipolar geometry, image matching algorithm based on the epipolargeometry constraint is presented. This algorithm can get the parallax of correspondingmatching points in two images through matching the corresponding feature of twoimages. By the basic principle of binocular stereo vision, goal body depth informationcan be calculated. By comparing the calculated depth information and practical depthinformation again, the accuracy of the depth information recovery based on the stereobinocular vision system is analyzed finally.