基于自适应聚合的立体视觉合作算法

Adaptive Aggregation Based Cooperative Stereo Vision

提出了一种恢复高质量稠密视差图的立体视觉合作算法.该算法采用基于形态学相似性的自适应加权方法,迭代地进行局部邻域的自适应聚合和抑制放大,实现高效率和高质量稠密视差图计算.将该算法推广到三目摄像机立体匹配系统中,通过重建摄像机坐标系实现图像校正,并根据连续性假设和唯一性假设,建立视差空间中的支持关系和三目摄像机之间的抑制关系.实验结果表明,三目立体合作算法能够得到精确的场景视差映射,并可以实现多基线方向的遮挡检测.该算法特别适用于由多个廉价摄像机组成的立体视觉系统,在几乎不增加软件和硬件资源的情况下,就可以得到高质量的稠密视差图.

This paper proposes a stereo vision cooperative algorithm for high quality dense disparity mapping. This algorithm iteratively performs the local adaptive aggregation and inhibitive magnification based on the morphologic similarity with adaptive weight, and generates high quality dense disparity map effectively. This paper also extends the cooperative algorithm to trinocular stereo vision system. By rebuilding the camera coordinate system, the trinocular images are rectified, and the support area and trinocular inhibition area are established in disparity space based on the continuity and uniqueness constrains. Experimental results show that the trinocular stereo vision cooperative algorithm can generate accurate real dense disparity maps, and the occlusions in multiple baseline directions can also be detected. This algorithm is especially suitable for stereo vision system with multiple cheap camera to realize high quality dense disparity mapping without more hardware and software.