一种基于双目视觉的快速前景提取方法

In computer vision, real-time foreground extraction is a challenging task to achieve, especially when dealing with video sequences. Almost all the related methods use captured image as the only input information. In order to get a better foreground area, the state-of-art algorithm often requires manual intervention or a complex model, which make it impossible to realize in practical application. In this article, we present an automatic approach to real-time bi-layer video segmentation using binocular camera device. ThisA Fast Approach to Bi-layer Segmentation of Binocular Video method combines the depth and edge information to optimize the edge feature and estimate the target contour by a local search method with Canny edge. Experiments show that this method has a very high time efficiency and good noise immunity.