基于前景检测的视差优化算法

Parallax optimization algorithm based on foreground detection

目前立体匹配算法分两类,传统的匹配算法通过计算两幅图的像素点相似程度,采用的是一种局部优先的方法。而当前的策略主要将问题转化为求解能量方程,进而对全局空间进行优化,提高匹配精度,获得更好的视差图。但是在实际应用过程中,由于光学失真和噪声,平滑表面的镜面反射,投影缩减,透视失真,低纹理和重复纹理等影响,导致误匹配或者找不到匹配点,从而得不到有效视差。然而在特定的场合,可以利用有限的有效视差,基于前景检测以及最小二乘法,优化得到较为完整的前景视差图。

Currently, there are two types of stereo matching algorithms, the traditional stereo matching algorithm use a local optimum method by calculating the similarity of the pixels in two images. The current strategy is to create its energy equation, and then to optimize the global space, improve matching accuracy and obtain better disparity map. However, in practical application process, the photometric distortions and noise, specular surfaces, foreshortening, transparent objects and repetitive or ambiguous patterns will lead to false match or no match point, which lead to noneffective parallax. However, in certain situations, you can optimize the limited effective parallax to get a more complete disparity map, based on foreground detection and least square method.