基于斜平面平滑优化的半全局立体匹配

Semi-global Matching Algorithm Based on Improved Slanted Plane Smoothing

基于斜平面平滑的半全局匹配算法虽然能很好地处理深度不连续区域,减少视差断裂现象,具有很好的视差平滑效果.但此方法计算量大,并且若输入的左右图像对不是理想校正的,最终得到的视差图中可能会出现黑色小方块.针对上述问题,本文提出了一种基于斜平面平滑优化的双目立体视觉算法,优化了视差斜平面拟合过程,提高了斜平面平滑的运算速度;另,通过对初始视差图中无效的视差值进行临近插值填充,很好地解决了视差图中的方块现象;同时考虑到同一分割块内的像素梯度值应比较接近,对图像分割时用到的能量函数增加梯度项,使分割区域更加合理.实验结果表明,本文算法能取得更好的视差图,主观效果得到改善,运算速度提高约40%.

The semi-global matching algorithm based on slanted plane smoothing can achieve good results in disparity discontinuous area.It can reduce disparity fracture phenomenon and makes the disparity image more smoothing.However,the slanted plane smoothing algorithm has huge computing quantity and long run time.If the input images are not very ideal,the final disparity image could have black small blocks.Aim at the above problems,binocular stereo vision algorithm base on improved slanted plane smoothing is proposed in this paper.The proposed algorithm not only increases the speed of calculation,but also fills the invalid disparity in original disparity image using the nearest interpolation algorithms.Considering that the gradients of the pixels in the same segmentation should be close,the gradient energy is supplemented to optimize the energy function.So the proposed algorithm segments the image more reasonably.The experiments demonstrate that the proposed algorithm has a better disparity result.The subjective quality is improved,and the time consuming is also increased by 40%.