Research On A Sparse Stereo Matching Algorithm Based On Belief Propagation

Stereo vision with a wide range of applications in three-dimensional measurement, robot vision, virtual reality and other fields is an important research field in computer vision. Stereo matching establishing the correspondence between features which selected is extremely difficult and key step in stereo vision, and its goal is to get disparity map.The global stereo matching is generally modeled as an energy minimization problem, it use a variety of optimization methods to find the minimum solutions of the energy function. There are many optimization methods such as dynamic programming, graph cuts, belief propagation and so on. Graph cut algorithm and belief propagation (BP) algorithm are better in global stereo matching. However, these algorithms computing time are math more than the local ones. This thesis proposed stereo matching algorithm is based on belief propagation algorithm, and is committed to reduce the algorithm computation time.Delaunay triangulation (DT) mesh is one form of image description. Delaunay triangulation algorithm using image feature points as vertices of triangles to generate an image content adaptive Delaunay triangle mesh. The image can be effectively restored by interpolation from the values of grid node.This thesis first introduces the stereo vision system, the basic principles and steps of stereo matching, and then analyzes the nature of DT mesh and its algorithms. By using the advantage of DT mesh in image coding field, there proposed a sparse stereo matching algorithm based on belief propagation. Algorithm is as follows:First, extract the features of the left image, to obtain the sparse left image, and use of these features gets a content adaptive mesh DT. Then use belief propagation-based stereo matching algorithm, match sparse left and right images to get sparse disparity map. These reduce the amount of time and in ensuring the accuracy of matching algorithms. Finally using the image reconstruction based DT mesh, an interpolation method can recover the dense disparity map from sparse one. The results obtained show that the proposed method obtains accurate disparity map, while reducing the matching time.