Research Of Stereo Matching Algorithm In Machine Vision

With the progress of society,people are expecting to lead more convenient and intelligent lives.Machine vision which acts as an important part ofdriverless vehicle sensing system receives more and more attention and research.Stereo vision,as an important field of machine vision,enjoys a wide prospect of application,since it can reconstruct three dimensional scenes and assist driverless vehicles to better perceive outside environment.However,there still exist two challenges: poor quality of disparity map and a large amount of computation,which affects practical applications of stereo vision.In order to handle these challenges,research of stereo matching is launched from two aspects: slanted surface and improper segmentation.The frontal-parallel assumption is made by many matching algorithms,which means that all planes in the scene are parallel to image plane,but this assumption fails for slanted surfaces.This thesis proposes a matching algorithm intending to improve the matching results for slanted surfaces,called Efficient Methods Using Slanted Support Windows for Slanted Surfaces.Firstly,a mathematical model is constructed to prove that slanted surfaces in the environment have corresponding slanted disparity surfaces in the disparity space image(DSI),and the model is to help find the proper plane parameters of slanted support windows.Then,improved cost aggregation and post-processing methods are proposed.The algorithm is tested using the Middlebury and Karlsruhe Institute of Technology and Toyota Technical Institute at Chicago(KITTI)benchmarks.The results demonstrate that the algorithm exhibits good performance and is efficient for slanted surfaces.Image segments are often used as a constraint in stereo matching.However,both over-segmentation and under-segmentation can lead to disparity degradation in some regions.In order to handle this challenge,the thesis proposes an algorithm: Stereo Matching Using Segment Optimization.To obtain an accurate disparity map,we introduce the notion of object and build a new global energy function to optimize segments and estimate a disparity map jointly.The effective and efficient block coordinate descent approach is used to optimize the global energy function by merging small segments to produce accurate disparity map.Meanwhile,segments can be optimized by merging similar regions,which helps to find a balance between over-segmentation and under-segmentation.We demonstrate performance of the proposed algorithm on KITTI benchmarks.The results show that our algorithm outperforms many state-of-the-art methods and confirm the effectiveness of approach.Besides,the occlusion and weak texture regions problems in the field of stereo matching are analyzed systematically by finding out reasons why these problems arise.After studying a lot of recent literatures,we provide a taxonomy of how to address these issues and make a summary for them,which can help other researchers to gain an overall understanding of occlusion.This thesis does the research on slanted surface and improper segmentation and analyzes the reasons.Two algorithms to handle the challenges are proposed.Study of occlusion and weak texture is also conducted.