The Study On Stereo Matching Of The Feature Line Based On Special Corner

Stereo matching is not only the key problem but also a bottleneck problem inmachine vision. The feature matching in stereo matching possesses a stronganti-interference ability, less computational effort, and relative faster processingspeed. It can meet the general needs. Although the feature matching can’t get denseparallax figure, it is of great significance in the study on pattern recognition, imageunderstanding, extraction and matching of image feature and other fields. Linefeatures have good set analyticity, and can easily describe goal. For binocular visionsystem, the accuracy and feasibility of extraction and matching of the line featuresdirectly influence the identification of the object.In real life, the outline of artificial things most can be described with segment.For example, outdoor building, roads, most indoor objects. So this paper mainly studythe quick and accurate extraction and matching problem of feature line.Aimed at the extraction of feature line, this paper presents an extraction methodof approximate feature line based on special angular point. The effective combinationof Hough and Harris algorithm can improve the speed of extracting segment usingHough algorithm and can get more accurate endpoints. The paper’s main thought iscoming from Harris algorithm on extraction of angular point. It has some advantages.On one hand, it possesses simple calculation, fast processing speed, and can extractlots of angular points. On the other hand, it can control the extraction number ofangular point, and can choose part of the point according to need. Then based on theextracting special angular point (angular point corner and large quantities of point), itcan use the Hough algorithm to extract feature line. But the Harris algorithm and theHough algorithm have their own disadvantages. Harris algorithm except a part ofnoise using the Gaussian filtering,but at the same time play a fuzzy and weakeningeffect on the fringe, so as to it can not extract lots of special angular point, and mayalso extract ruleless angular points. Hough algorithm can’t accurately get the endpoint.This paper offers two methods to solve these problems. First, use sharpening methodto extrude the edge in the Harris algorithm, and then reuse texture density measurement operator filter textures angular point, so as to reduce the pixel motifs. Itcan improve the speed of extraction segment. Second, extract the endpoint of thesegment accurately as far as possible, so as to overcome the Hough algorithm’sproblem of cannot precisely getting endpoints.Aimed at the matching of feature line, this paper reduce search space usingangular point constraint, and then use the matching method of combination of colorinformation correlation matching and gray matching, so as to improve the precisionof matching. Traditionally it is using limiting constraint to make search space fromtwo-dimension to one-dimension. This paper puts forward angular point restraint. Itcan reduce the search space to several points, and through combining the advantagesof color matching algorithm to make up for the deficiency of the gray matching, likethe light sensitivity, so as to improve the precision of matching. And further use thematching point that had been found by the least squares estimate to form matchingsegment, so as to improve the success rate of matching.