基于区域插值的视差图平滑算法及三维重建

Disparity Map Smoothing Algorithm and Three-dimensional Reconstruction Based on Region Interpolation

立体匹配作为双目立体视觉的关键步骤,易受遮挡和光照等因素影响出现误匹配,得到包含很多亮点的不平滑视差图,进而影响恢复物体的三维结构。为了消除误匹配区域、平滑视差图,文中提出采用基于区域的插值算法和三角网格线性插值算法对视差图进行后处理。首先采用Middlebury标准立体数据集中的图像对进行立体匹配获得视差图,其次采用基于区域的插值算法对视差图进行插值,设定阈值剔除误匹配点对,然后通过三角网格线性插值算法对误匹配区域进行填充,最后根据平滑处理后的视差图进行三维重建。实验结果表明本文所提方法能够较好的恢复物体的三维结构,同时减少了计算量、提高了处理效率。

Stereo matching is a key step in binocular stereo vision.It is susceptible to erroneous matching due to occlusion and illumination,and obtains a non-smooth disparity map containing many bright points,which in turn affects the three-dimensional structure of the restored object.In order to eliminate the mismatched area and smooth the disparity map,in this paper,the region-based interpolation algorithm and the triangular mesh linear interpolation algorithm are proposed to post-process the disparity map.Firstly,stereo matching is performed using the image pairs in the Middlebury standard stereo dataset to obtain a disparity map.Secondly,the disparity map is interpolated by employing the region-based interpolation algorithm,and set threshold exclude mismatched point pair.Then,the mismatched area is filled by using a triangular mesh linear interpolation algorithm.Finally,three-dimensional reconstruction is performed according to the smoothed disparity map.The experimental results show that the proposed method can better,recover the three-dimensional structure of the object while reducing the calculation cost and improving the processing efficiency.