Research On Matching Techniques Of Stereo Vision

Stereo vision has traditionally been one of the most extensively investigated topics in computer vision.Its goal is to make the computer have the ability of gaining 3D information from 2D images.It has been widely used in three-dimension measurement,robot navigation, virtual reality and so on.Firstly,the principle of stereo vision system is introduced in detail in this paper.Then,the paper focuses on the technique of camera calibration and stereo matching.Camera calibration is a necessary step in stereo vision in order to extract metric information from 2D images.It is the process of determining the internal camera geometric and optical characteristics and the 3D position and orientation of the camera frame relative to a certain world coordinate system.A binocular stereo vision system is constructed in lab and calibrated using Zhang's method based on planar calibration board.Then,the thesis gives its measurement decision by measuring the length of the board grid.Stereo matching is one of the most active research areas in computer vision.It is also the key and the most difficult problem in stereo vision.The correspondence problem is naturally stated in terms of energy minimization.Once the global energy has been defined,a variety of algorithms can be used to fred a minimum,including dynamic programming,graph cut,belief propagation and so on.By far,the most common approach to global matching is dynamic programming,which uses smoothness constraints to optimize correspondences in each scan-line by finding the minimum cost path through a disparity space image,ignoring the interscanline constraints.To solve the problem,the paper presents a stereo matching algorithm based on dynamic programming in two dimensions.Firstly,compute the primitive disparity map using dynamic programming in the row direction.Then based on the results,give the corresponding data term a reward in order to be used by next dynamic programming process.Lastly,produce the final results using dynamic programming in the column direction.Recently,belief propagation is becoming more and more popular in stereo matching based on energy minimization.After introducing the principle of belief propagation,this thesis gives a stereo matching algorithm based on belief propagation and color segmentation. Firstly,compute the left and right disparities using belief propagation.Secondly,stereo match consistency is used to detect occlusion regions.Lastly,the scene structure is modeled by a set of planar disparity planes based on the results of color segmentation and the disparities of the oeclusion regions is recomputed.Experimental results demonstrate that the algorithm has good performance.