提高霍夫变换识别圆形物体准确率的算法

An algorithm for improving the precision of detecting round object by Hough Transform

霍夫变换(Hough变换)有对不完整边缘具有鲁棒性的优点,然而,这个优点在有些时候却会导致目标的错误识别。大量实践表明:对实际圆形物体实施Hough变换检测,检测到的点通常集中在某一段比较连续的圆弧,如果检测到的点分散在圆的各个位置,则通常是噪声的干扰造成了目标的错误识别。本文在此基础上提出了一种算法,即对被检测到的边缘点的角度求标准差,根据其标准差值与被检测到的边缘点数进行比较,以此来判断该圆形物体的识别是否正确。本文将该算法应用于人头识别,取得了良好的效果。

Hough transform has the advantage of detecting discrete edges. But the advantage sometimes leads to wrong recognition. A lot of practice indicates that when detecting round objects by Hough transform, the dots detected always focus on some segment of continuous arc. If the dots detected are dispersed on the circle, it would be a wrong recognition caused by noise. Based on this point, this paper raises an algorithm to solve it. That is to get the standard variance of the angles of dots detected, and comparing the standard deviation with the number of edge dots detected to judge whether the recognition is right or not. In this paper, the algorithm is put into use of detecting persons＇ heads, and get good effect.