半全局立体匹配算法的改进研究

针对半全局立体匹配中单一 Census 精度不足且边缘区域视差效果较差的问题,提出一种基于颜色差信息 BTCensus 和加入分割约束的半全局匹配算法。首先该算法采用联合三通道 BT 算法和 Census 变换的计算方式实现代价计算,缓解重复纹理歧义并提高不连续区域配精度;同时将图像分割得到的场景信息作为约束对代价聚合函数进行改进,保证割块内部的纹理平滑,提高边界纹理区域的匹配精度;最后引入图像分割信息进行分步中值滤波,平滑视差图。实验表明,该算法在重复纹理区域、视差不连续区域和弱纹理区域都取得了较好的视差效果,可获得高匹配率的视差图。

Improvement of Semi-global Stereo Matching Algorithm

A semi-global matching algorithm based on color difference information BTCensus and segmentation constraints is proposed to solve the problem of poor precision of single Census and poor disparity in edge region.Firstly , the algorithm uses the combined three-channel BT algorithm and Census transform to realize the cost calculation , which can reduce the repeated texture ambiguity and improve the matching precision of discontinuous regions.At the same time , the scene information obtained from image segmentation is used as a constraint to improve the cost aggregation function to guarantee the texture smoothness and improve the matching precision.Finally , image segmentation information is introduced to filter the median step and smooth the disparity map.The experimental results show that the algorithm has good disparity effect in repeat texture area , disparity discontinuity area and weak texture area , and can get disparity map with high matching rate.