Research On Stereo Matching Algorithm And Its Application Based On Image Segmentation

Binocular stereo vision system is an important research field in the field of computer vision.It has a wide range of applications in three-dimensional measurement,robot navigation,intelligent control,medical imaging,virtual reality and so on.In fact,Binocular stereo vision can be understood as imaging the same object or scene from two different perspectives,and then calculating the spatial distance corresponding to the matched pixel points in the two images so as to finally restore the three-dimensional information of the object or the scene.A complete binocular stereo vision system consists of four parts: image acquisition,image stereo correction,stereo matching and 3D reconstruction.Among them,stereo matching is one of the core problems in binocular stereo vision.Traditional stereo matching methods include local stereo matching methods and global stereo matching methods.However,the traditional stereo matching method uses stereo pixels as the basis for stereo matching.The stereo matching method based on pixel points has two disadvantages.On the one hand,the matching accuracy of the low texture region,the occlusion region and the disparity discontinuous region is low.On the other hand,the calculation is complex and the timeconsuming is long.For these two aspects of the problem,based on researching the local stereo matching methods,the global stereo matching methods,image superpixel segmentation methods and global stereo matching methods based on image segmentation,we propose the LBP stereo matching algorithm based on the WKK-SLIC superpixel segmentation.The main innovations and research results are as follows:(1)Proposing the stereo matching method based on image segmentation.The traditional stereo matching methods are divided into a local stereo matching method and a global stereo matching method.The local stereo matching method has fast matching speed but the accuracy is low.The global stereo matching method has high accuracy but takes a long time.In this paper,a new stereo matching method based on superpixel segmentation is proposed for a specific application scenario.This method guarantees high accuracy while the algorithm has a low time complexity.(2)For image segmentation algorithm,proposing a new SLIC superpixel segmentation algorithm(WKK-SLIC algorithm).We verify the algorithm on the Berkeley Segmentation Database public data set.The experimental results show that the WKK-SLIC algorithm is superior to the SLIC algorithm in evaluating several common standards for superpixel segmentation.(3)For the global stereo matching based on image segmentation,based on the WKK-SLIC superpixel segmentation algorithm,proposing the LBP stereo matching method based on the WKK-SLIC superpixel segmentation algorithm.We test the algorithm on the Middlebury test platform.The experimental results show that the stereo matching method proposed in this paper not only has high matching accuracy,but also can effectively handle the matching problem of the low texture area and the occlusion area.