论文部分内容阅读
本文提出一种新的能快速从 2幅投影图像中自动检测平面的方法 .首先通过引入互相关对Scatt特征点匹配结果作后处理得到可靠的点点匹配 ,然后根据 2幅图片间特征分布的近似不变性从特征点匹配得到特征线匹配 ,继而将特征点按照与特征线的几何关系进行分组 ,作为初始集合估计平面单应矩阵 ,将特征点按所属平面分组 .此方法能有效地避免随机的全局搜索 .此方法尤其适于人造场景的平面检测问题 .实验表明了此方法的有效性 .
In this paper, we propose a new method that can automatically detect the plane from two projection images.Firstly, through the introduction of cross-correlation, the matching results of Scatt feature points are post-processed to obtain a reliable point matching, and then based on the approximation of the feature distribution between the two images Invariance The feature points are matched from the feature points, then the feature points are grouped according to the geometric relationship with the feature lines, which is used as the initial set to estimate the plane homography matrix, and the feature points are grouped according to their planes. This method can effectively avoid random This method is especially suitable for the plane detection of man-made scenes.The experiments show the effectiveness of this method.