| Recently, the stereo vision is widely used in the military, medical and industrialinspection field, and it also plays an important role in the vision measurement. The stereomatching technology is the most important and difficult part in the stereo vision, and itsresult decides the accuracy of the whole measurement results in the vision measurement.The project focus on the research of the stereo matching technology from several aspects,such as the matching rate, the elapsed time and the accuracy.In order to resolve the defects that the traditional Harris comer detector does not havethe invariant property and be sensitive to the noise, an improved multi-scale Harris cornerdetection algorithm was proposed. Aiming at the accuracy of the homograph, an improvedRansac algorithm was presented. According to the problems of matching rate andaccuracy, a method based on the SURF and Harris algorithm was proposed, in which theHarris algorithm has the feature of scale invariant.Firstly, the background and significance of the stereo vision in the visionmeasurement and the research status of the stereo matching technology were introduced,and some common stereo matching methods were analyzed.Secondly, some basic feature extraction and feature matching algorithms wereresearched, and each algorithm above was tested with the same image under the VisualC++ platform.Thirdly, the principles and steps of the improved algorithms provided in this projectwere summarized, such as the improved multi-scale Harris corner detection, the improvedRansac algorithm and the method based on the SURF and Harris algorithm.At last, the ZhangZhengYou calibration method was usded to calibrate the binocularstereo vision cameras, and the data of calibration is considered as the reference of theexperiment results. And the epipolar geometry model was used to check the accuracy ofall the methods provided in this paper. The experimental results show that this method canreach a good condition of feature detection and matching. |
PDF Full Text Request