区域分割的亚像素相位立体匹配算法

A Stereo Matching Algorithm of Sub-Pixel Phases Based on Region Segmention

针对双目光栅投影测量中深度不连续及边界阴影处的误匹配导致测量误差大的问题,提出区域分割的亚像素相位立体匹配算法。该算法采用由粗到精的匹配策略,在标准极线几何、相位等约束下获得粗匹配点。根据变形栅线图、展开相位图的特点将粗匹配结果分为不包含边界阴影的深度连续区域、边界阴影及深度不连续区域,分别使用加窗匹配测度最小二乘法和外插法获得亚像素匹配点。实验结果表明:该算法能有效消除传统匹配算法在深度不连续处的视差平滑过渡;与传统点基元匹配算法相比,该算法在边界阴影及深度不连续处的平均高度相对误差减小了62%~69%,与传统模板匹配算法相比,该算法在边界阴影及深度不连续处的平均高度相对误差减小了57%~61%,在深度连续处的匹配精度提高了3~4倍。该算法有效提高了相位立体匹配的精度,对提高目前三维测量技术精度具有重要意义。

A stereo matching algorithm of sub-pixel phases based on region segmentation is proposed to solve the problem of mismatching in boundary shadow region and depthdiscontinuous region appeared in binocular projected grating measurement. The proposed algorithm uses a matching strategy for coarse-to-fine and obtains the coarse matching points under a standard epipolar line constraint,aphase constraint and other constraints.The algorithm divides the coarse matching results into a depth-continuous region without boundary shadow and a depth-discontinuous region with boundary shadow in accordance with the characteristics of the deformed fringe map and the phase unwrapping map.Sub-pixel matching points are then obtained by using the least square curve fitting method with windowed matching measure and the extrapolation method,respectively.Experimental results show that the proposed algorithm effectively eliminates the disparity smoothing transition in depth-discontinuous region. A comparison with the traditional point matching algorithm and traditional template matching algorithm shows that the proposed algorithm reduces mean relative error of heights from 62% to 69% and from 57% to 61%,respectively in depth-discontinuous region and increases thematching accuracy 3 to 4 times in depth-continuous region.The proposed algorithm effectively improves the accuracy of phase stereo matching,and is of great significance to improve the accuracy of 3D present measurement technique.