一种基于改进Hough变换的直线快速检测算法

Fast line detection algorithm based on improved Hough transformation

针对传统Hough变换计算量大、耗费内存空间、参数空间峰值点被次峰值点包围、易造成漏检或误检等缺陷,提出一种改进的Hough变换直线快速检测算法。首先检测图像中相邻的像素点并进行聚类,形成一些相连的像素点的集合,然后将聚类后的像素点进行感知编组,细分成比原聚类线段更接近直线的线段,最后对每段近似直线用随机Hough变换进行检测,从而精确地检测出图像中相应的直线。实验表明,与传统Hough变换相比,改进后的算法计算量小,节省内存,无需先验知识,且抗干扰性有显著提高,并降低了误检率和漏检率。

Hough transform（HT） is a popular tool for line detection due to its robustness to noise and missing data.However,the computational cost and memory space consumption associating with its voting scheme have prevented its applications.Here an improved HT algorithm is proposed to solve these problems.Firstly the neighbor pixels are clustered,and then the clusters are subdivided into sets of most perceptually significant straight line segments.For each segment,its best fitting line can be found using random Hough transform（RHT）.Compared with traditional HT algorithm,the proposed approach can not only accelerate the computing speed and save memory space,but also produce a much cleaner voting map and make the transform more robust.