结合图像分割的半全局立体匹配算法

Semi Global Stereo Matching Algorithm Combined with Image Segmentation

针对半全局立体匹配(semi global matching,SGM)算法对于视差突变的像素给定一个统一的惩罚,会导致边界的模糊以及弱纹理、遮挡区域误匹配的问题,提出加入图像分割的SGM算法。该算法首先在代价计算过程中采用灰度绝对差(absolute difference,AD)变换,保证实时性;然后在代价聚合部分采用改进的SGM算法,使算法具有较强的鲁棒性及较高精确度的同时,降低了在弱纹理、遮挡区域的误匹配率;在视差选择部分,通过抛物线拟合法得到亚像素级分辨率视差;在视差优化步骤中,利用左右一致性检测对遮挡点进行检测,并在此基础上对无效视差值进行赋值填充;最后使用中值滤波进行优化,确保影像的清晰度和完整性。实验结果表明,该算法可有效降低在弱纹理及遮挡区域的误匹配率,满足立体匹配的需求。

Aiming at semi-global stereo matching algorithm,a uniform penalty is given for the pixels with disparity mutation,which leads to the blurring of the boundary and the mismatch of weak texture and the region of occlusion.A semi-global stereo matching algorithm with image segmentation information is proposed.Firstly,absolute difference(AD)transformation is used by the part of cost calculation process to guarantee real-time in this algorithm.Then,in the cost aggregation part,the improved SGM method makes the algorithm more robust and more accurate.Next,subpixel resolution disparity is obtained by parabola fitting in the part of disparity selection part.Finally,the initial disparity map can be obtained and its occlusion point is detected by the left and right consistency test,and the invalid points are evaluated and filled on the basis of this.Median filtering is used by disparity map optimization to ensure the clarity and integrity of the image.Experimental results show that the algorithm can effectively reduce the mismatch rate in weak textures and occlusion regions,while meeting the requirements of stereo matching.