基于匹配点相似度引导采样的图像多平面检测

Multi-planar detection for digital images based on guided sampling of matching point similarity

针对实际广泛存在的具有深度变化的多平面场景,提出一种基于匹配点引导采样的多平面检测算法。根据平面结构中匹配点的相似度改进匹配点的采样规则,通过对匹配点的残差信息进行排序和加权分析,获取场景中各个平面结构对应的单应矩阵,从而实现了场景中多平面区域的准确检测。实验结果表明,与现有典型的算法相比,提出的算法可获得较高的平面检测准确率。

This paper proposed a multi-planar detection algorithm based on guided sampling of matching point similarity for the widely existing multi-planar scenes with changing depths. According to the similarity of matching points, it improved the sampling rule. By sorting and weighted analyzing for the residual information, the homographic matrix corresponding to each planar struc- ture could be obtained, and all the planar structures could be detected successfully. Experimental results demonstrate that, com- pared with the existing typical algorithms, the proposed algorithm can achieve higher accuracy of the multi-planar detection.