Research On Key Technology Of Binocular Stereo Vision In Three-Dimensional Reconstruction

Computer vision is a scientific research on how to use computer to realizethe function of human vision, more precisely, it is using camera and computerinstead of human eye to complete perception, recognition and understanding ofthe objective three-dimensional scene. Binocular stereo vision is a significantbranch of computer vision, which can directly simulate human eye to deal withthe objective scene. It has characteristics of low cost, high efficiency and highflexibility, which is widely used in industry, agriculture, aerospace, medicaltechnology and other fields.In the application background of3d reconstruction based on binocular stereovision, based on the hardware foundation of self-built binocular stereo visionplatform, key techniques in binocular stereo vision including camera calibration,image preprocessing, stereo correspondence and3D reconstruction are exploredand researched in depth. The prime research task content is as follows:1. In camera calibration section, Zhang plane calibration method is adopted.Complete stereo calibration and calibration of intrinsic and external parameters ofthe left and the right camera using calibration toolbox.2. In image preprocessing section, use adaptive median filtering method toremove noise distortion introduced in the process of imaging. Use contrastlimited adaptive histogram equalization method to enhance the image contrast,and balance the brightness difference between left and right images. This sectionlays the foundation for subsequent stereo correspondence.3. In stereo matching section, an improved two-step stereo correspondencealgorithm based on combination of feature-matching and region-matching isproposed. Use Harris algorithm to extract the feature corners, estimate disparitysearch range through corner matching, and then use the calculated disparity search space to proceed region matching. In region-matching step, error thresholdmechanism is introduced to filter the unbelievable disparity estimation point.Finally, the overall dense disparity map is obtained.4. In3D reconstruction section, according to the basic principle of disparitydistance measuring method, adopt triangulation method to calculate the obtaineddisparity map, then the depth map and the3D display figure are received.Complete three-dimensional reconstruction of standard Tsukuba images and realscene images collected by actual system.