Image Rectification Of Stereo Pairs And Depth Measurement Research

The measurement of the depth information is a very important issue. As a significantmeasuring technique of passive vision, the stereo vision is highly paid attention to in theresearch of computer vision. The traditional binocular measurement method first needs to usecamera calibration to obtain the camera intrinsic and extrinsic parameters. Then it uses theimage rectification with calibrated parameters to obtain the image without vertical disparity,which can speed up the stereo match. At last, it uses the triangle principle to calculate thedepth and the length of the corresponding points.Based on the stereo vision, this paper mainly studies the binocular camera calibrationmethod, the rejection of the false match in the corresponding points, and the imagerectification with the weakly calibrated information. The camera calibration can be equivalentwith setting up two tables, and simplifying to estimate the camera parameters. Using thelocation of the corresponding points in stereo images, which contains the weakly calibratedinformation, can rectify the original images to the images only with horizontal disparity, andto minimize the image distortion.The main contents of this paper are:Firstly, this paper studies the camera calibration, the relationship between the parallaxand depth measurement in stereo vision. According to some optimizing process, it canimprove the measuring accuracy and use the parameter estimation to simplify two setup tables.Secondly, in addition to the random sampling consensus method, this paper uses theclustering paring consensus method to reject the false match in the photos with non-planarscene and improves the accuracy.Finally, this paper compares different weakly calibrated image rectification methods.After analyzing the reason of the image distortion in weakly calibrated image rectification, itcan use the pyramid search and more appropriate cost function to improve the accuracy ofimage rectification, and use the gene algorithms to shorten the average searching times.