场景轮廓的动态规划立体匹配算法

Dynamic programming algorithm for stereo correspondence of contour

立体匹配算法是三维重建的关键步骤。由于实际场景中经常存在大片灰度相近的区域,稠密三维重建存在计算时间长、实时性差的问题。采用了场景的轮廓来重建场景的方法。基于场景中相邻点之间的视差应当是连续的假设,解决了轮廓在匹配时存在的"噪点"的问题,利用动态规划法对轮廓上各个点的视差进行约束以及求解最优解。由于提取轮廓后需要匹配的点数大为减少,用时可减少为原来的10%,得到与场景一致的轮廓视差图。

Stereo correspondence is one of the most promising research areas in stereo vision. In order to solve the problem that there are large areas of similar gray-scale region and the time-consuming of the dense reconstruction, the contour of stereo pairs is extracted to reconstruct the scene. To reduce the noise which may lead to incorrect results during the stereo matching, a stereo matching algorithm based on the dynamic programming is proposed. Based on the reasonable and commonly assumption that the disparity of the scene is always smooth, the stereo matching algorithm puts a restraint on the disparity map and applies the dynamic algorithm on the DSI to obtain the optimal disparity of the contour. Since the sum of points needed to be computed is greatly reduced, the algorithm can implement the stereo correspondence with high speed and matching rate.