Stereo Matching Algorithm With Modified Non-parametric Transform Measure

Stereo matching is a basic algorithm and key problem in vision measurement, itsaccuracy directly determines the precision of vision measurement. The similarity measureis the degree value which is used to measure the similarity among the image feature uints,it is an important factor and control condition in stereo matching. Therefore, activelyresearch on the similarity measure is very important for improving the precision of thevision measurement precision and matching technology.Based on the study of stereo matching, the similarity measure is introduced to thematching algorithm based on region and feature, and its important effect is verified instereo matching. A strong anti-interference modified similarity measure is proposed.In order to make the matching algorithm to satisfy the requirements of high precisionand anti-interference, this paper presents a non-parametric Census transform measurebased on local texture measurement to its limitation. Combining with semi-globalmatching algorithm to aggregate cost. By using image segmentation method to refine theinitial results and the dense disparity map is obtained. Several experiments and evaluationdatas show that, the modified Census measure not only significantly improves thematching accuracy, but also can overcomes the influence of amplitude distortion in thematching effectively.Based on the multi-view geometry relationship, the homography constraint ofplane-sweeping method is proposed and applied to multiple view feature matchingalgorithm. Because the low accuracy of homography problem, introducing similaritymeasure to obtain accurate matching points and eliminate mistakes of common constraintsresulting in occlusion and the characteristics of intensive regions. The rationality of thealgorithm is verified by experiment and it can be applied to3D vision measurement.