基于视差图像序列的深度估计算法研究

Research on Depth Estimation Algorithm Based on Parallax Image Sequence

现有的深度估计算法中,Kim算法不需要全局优化,可以保留精确的目标轮廓,同时仍然确保在少量细节的区域中平滑重建,得到的效果相对较好,但是其在深度扩散算法中时间复杂度太高,不适合广泛应用.针对这一问题,我们对其深度扩散算法进行改进.首先,根据边缘图像的深度范围覆盖整个场景的深度,将深度扩散的方式转变为在边缘深度范围内的快速搜索而非整个区域的全局搜索,明显降低了算法的时间复杂度.另外,根据极平面图像(EPI)的线性结构,将深度扩散沿着斜率的方向对多个视角进行综合考虑扩散,而非只是针对单个视角下的像素点进行,使非边缘平滑区域的扩散更加精准.经过仿真验证,本文的算法在一定程度上优于Kim算法,可以广泛使用.

As one of the existing estimation algorithms,Kim algorithm does not need global optimization.It can still ensure smooth reconstruction in a small area of detail while keeping accurate contour of the target.The results are relatively good,but the time complexity is too high in the depth diffusion algorithms,not suitable for wide application.In order to solve this problem,the depth diffusion algorithm is improved.Firstly,according to the depth of edge images,the entire scene depth is covered,and the method of deep diffusion is transformed into a quick search within the edge depth,rather than a global search for the whole area.,significantly reduces the time complexity of the algorithm.In addition,according to the linear structure of epipolar-plane image,the depth diffusion is taken into consideration with multiple views along the slope,rather than just pixels in a single view,which makes the diffusion of non-edge smooth areas more accurate.The simulation results show that the algorithm we proposed in this paper is better than the Kim algorithm to some extent and can be widely used.