立体图像的遮挡边界区域检测技术

A detection technique of occluded boundary regions for stereo image pairs

针对平行结构立体摄像机存在较大的遮挡边界区域问题,本文提出了一种基于可信图的遮挡边界检测算法。首先通过采样计算立体图像对中几条扫描线的可信值,从而得到遮挡边界的若干采样点;然后通过RANSAC算法得到遮挡边界。理论和实验分析表明该算法对复杂背景的立体图像有好的检测精度;对人肩像等简单背景图像的检测精度不好,但通过校正后仍可得到较准确的遮挡边界。

Aimed at the presence of large occluded regions in stereo image pairs obtained from stereo camera pairs, an algorithm which can detect occluded boundary areas based on reliability map is proposed. Firstly, a number of sampled occluded boundary points are obtained by calculating reliability of the disparity on the corresponding sampled scan lines. Secondly, the boundary llne is calculated by RANSAC method. Theoretical and experimental analyses showed that the proposed algorithm gives a good boundary detection accuracy for stereo images with complex backgrounds. But for stereo images with homogenous background, the results are not good, however, the algorithm can also give a relatively good accuracy after certain correction.