基于圆环模板和标志矩阵的角点检测算法

A Corner Detection Algorithm Based on Ring Template and Mark Matrix

为了解决计算机视觉系统中图像边缘检测时边缘点和边缘段同时检出的问题,提出了一种基于圆环模板和角点标志矩阵表示的角点检测算法。首先通过设置较小的阈值,应用圆环模板提取较多的灰度图像的亮度变化点作为候选角点;记录角点在原图像中的位置,并将原图像中该像素值设置为1,其他像素设置为0,建立角点标志矩阵,并将矩阵按照四叉树结构分解;统计子矩阵的1的个数和所占比例,将个数和比例较低的子块的角点标记删除;对其他的子矩阵分别采用行方式和列方式进行行程编码,在编码中对连续的1序列仅保留第一个,将后续的1置为0。记录同时在行方式和列方式结果矩阵中的1元素为角点。实验结果表明,与Moravec算法、Harris算法与FAST算法相比,该算法具有角点检测准确和适用面广的优点,在一定程度上解决了减少角点数目的问题。

In order to solve the problem of simultaneous detection of edge points and edge line segments during the edge detection in computer vision system,we propose a corner detection algorithm on basis of the ring template and the corner mark matrix. At first the smaller threshold is set up. More pixels with the brightness change extracting by the ring template are treated as the candidate corner pixels. The corner locations are recorded in the original image,and the pixel value of the original image is set to 1,and the other pixels are set to 0. The corner marker matrix is established and decomposed according to the quadtree structure. The number and proportion of the 1 are calculated in the submatrices. And the corners in the subblocks are deleted from the corner marks with the lower numbers and the lower proportions. The sub matrices are coded in line and column respectively through run length coding. The first 1 element in the consecutive coding sequences is kept,and the subsequent 1 is set to 0. The 1 element in the result matrix of the row and column mode are recorded as the corner points at the same time. The experiment shows that compared with Moravec algorithm,Harris algorithm and FAST algorithm,the proposed algorithm has the advantages of the accurate corner detection and the wide application,which solves the problem of reducing the number of corner points to a certain extent.