摘要
为降低双目立体匹配算法在视差不连续区域和噪声干扰情况下的误匹配率,提出了一种基于改进Census变换和动态规划的立体匹配算法。采用支持区域为十字交叉形状窗口且设有噪声容限的改进Census变换进行代价计算,提高了单像素匹配代价的可靠性;利用引导图滤波器快速有效地完成代价聚合;在视差选择阶段,设计了一种改进的动态规划算法,消除了扫描线效应,提高了匹配速度和正确率;经过视差后处理得到最终视差图。实验结果表明,该算法在Middlebury测试平台上的平均误匹配率为5.31%,在低纹理区域和视差不连续区域均能得到准确的视差,运算复杂度低且具有较好的稳健性。
In order to reduce the mismatching rate of binocular stereo matching algorithm in the disparity discontinuity region and under noise disturbance, a stereo matching algorithm based on improved Census transform and dynamic programming is proposed. An improved Census transform with a noise margin is applied to compute the cost based on a cross shape support region. The reliability of single pixel matching cost is enhanced. The guided image filter is used to aggregate the cost volume fast and efficiently. In the disparity selecting step, an improved dynamic programming algorithm is designed to eliminate the scan-line effect and improve the matching speed and accuracy. The final disparity maps are gained after post-processing. The experimental results demonstrate that the proposed algorithm evaluated on the Middlebury benchmark achieves an average error rate of 5.3194, and the accurate disparity can be obtained in both low texture and disparity discontinuity regions with low computing complexity and strong robustness.
出处
《光学学报》
EI
CAS
CSCD
北大核心
2016年第4期208-216,共9页
Acta Optica Sinica
基金
国家自然科学基金(61375025
61075011
60675018)
关键词
机器视觉
立体匹配
Census变换
动态规划
引导图滤波
视差
machine vision
stereo matching
Census transform
dynamic programming
guided image filtering
disparity
