Research On Adaptive Support-Weight Stereo Matching Algorithm

Stereo matching algorithm based on binocular vision is to find the matching pixels between left and right image with same scene.As a hot area of research in the computer vision field,stereo matching algorithm can be mainly divided into local and non-local algorithms,which have their distinctive advantages.Local algorithms have high matching efficiency and the other have high matching accuracy.This dissertation is devoted to improve the matching efficiency of adaptive support-weight(ASW)algorithm to make it better used to actual applications.Firstly,this dissertation introduces the research background and significance of stereo matching based on binocular vision,summarizes the development process and present research.Then we detail the constraints,four steps and evaluation criterions of stereo matching algorithms.At last,we introduce the basic principle and realization of the guided filter-based algorithm,Variable Cross algorithm and ASW stereo matching algorithm.Secondly,this dissertation proposes a redundant pixel removal mechanism to improve matching accuracy.First,we transform the left and right image to logRGB color space to remove the influence of environmental factor.In the selected support window,the weight factor is utilized to remove redundant pixels,and then reduce their influence on the matching accuracy.Then,we further select useful pixels to preserve the edge of objects using image corrosion.Only the remained pixels in aggregation window are used for cost aggregation.Moreover,WTA rule and left right consistency check is used to improve precision and then obtain a relatively accurate disparity image.Finally,a fast ASW stereo matching algorithm is proposed to reduce the computation complexity of traditional algorithms.First,a novel weight coefficient,which adapts cosine function to satisfy the weight distribution discipline,is proposed to accomplish the original cost aggregation.And then,the disparity range is divided into several sub-ranges,from them the local optimal disparities are selected.For each pixel,only the ASW at the location of local optimal disparities is calculated,thus the complexity of the algorithm is greatly reduced.Experimental results with a number of images show that the proposed algorithm can effectively reduce the amount of computation,as well as improve the matching accuracy to some extent,and achieve ideal results.