基于均值偏移的自适应区域增长视差图优化 - 计算机应用研究.doc

基于均值偏移和区域映射的视差图优化 张惊雷1,2,吉 玲2 (1天津市复杂系统控制理论及应用重点实验室,天津 300384；2天津理工大学 自动化学院,天津 300384) 摘 要：提出一种基于均值偏移算法。首先采用均值偏移算法对视差图进行区域分割,提取由误匹配所导致的黑洞区域。将左原始图像的分割区域映射到视差图中,从而对视差图进行每个区域内的平面拟合,去除误匹配点和像素跳跃点。然后结合标定数据得到三维场景的深度信息和三维坐标,最终实现三维重建。通过对图割算法及四种经典匹配算法进行实验算法能显著提高视差图的精度,降低低纹理区域和遮挡区域的误匹配率,且改善了重建效果。 关键词：均值偏移；视差优化；区域映射；三维重建Disparity map refinement based on image region segmentation and region mapping Zhang Jinglei1,2 , Ji Ling2 (1 Tianjin Key Laboratory for Control Theory Applications in Complicated Systems, Tianjin 300384;2 School of Electrical Engineering, Tianjin University of Technology, Tianjin 300384) 【Abstract】To solve the problem of low accuracy for initial disparity calculated by traditional matching algorithms, this paper proposed a disparity map refinement algorithm based on image region segmentation and region mapping. First the method segmented the initial disparity map into regions by mean shift algorithm, and then extracted the black holes caused by error matching. To remove these mismatching points and disparity jumping points, the algorithm mapped segmentation regions of the original image into the disparity map, and fitted planes with planar fitting algorithm. It calculated depth information and three-dimensional coordinates according to the calibration data. Finally it completed the reconstruction. This paper compared the refined disparity maps with the results calculated by the Graph Cut algorithm and other four traditional algorithms. Experimental results show that our method can greatly reduce the error matching rates, especially in textureless regions and occluded regions, and can improve the reconstruction quality. 【Key words】mean shift; disparity map refinement; region mapping; three-dimensional reconstruction 0 引言 基于立体视觉的三维重建一直以来都是机器视觉领域的研究重点,在视觉测量、交通导航、工业生产等领域都有着广泛的应用。双目立体视觉处理过程一般包括图像获取与预处理、摄像机标定、立体匹配和三维重建。三维重建主要受视差图和摄像机标定所得内外参数的影响,由于摄像机标定算法已经成熟,因此精确的视差图是三维重建的关键。传统的视差图计算分图像预处理和立体匹配两部分[1,2],由于同一景物在左右图像对中可能有较大差异,而且场景中存在光照、景物几何形状、纹理、噪声和畸变等复杂因素,且匹配过程中存在