基于RHT-LSM直线检测方法的研究

Method for line detection based on RHT-LSM

本文结合随机Hough变换(RHT)抗噪声能力强与最小二乘法(LSM)拟合精度高的特性,提出了一种基于随机Hough变换与最小二乘法进行直线检测的方法。该方法能用于背景噪声较强,直线存在一定弯曲的图像,检测精度高。首先,用随机Hough变换确定直线的大致位置,得到直线参量和数量;然后,利用所得直线参数,计算图像中的点到直线的距离,根据距离,可以确定每条直线附近的点集,剔除干扰点和噪声;最后,用最小二乘法对点集中的各点进行拟合,得到精确的直线参量。把该方法应用于列车动态识别中的制动梁检测,得到了良好的效果。

Based on strong resistance to noise of Randomized Hough Transform （RHT） and high fitting precision of Least Square Method （LSM）, a new method to detect lines in image was presented on the basis of RHT and LSM. At first, approximate position of lines was confirmed, and the parameters and quantity of the lines were computed by RHT. And then, the distance was computed from every point to each line detected by RHT. Based on the distances, the points on the line and near the line belonged to the same cluster, so several point sets could be obtained whose number was equal to the number of the line. The point sets in the vicinity of each line were extracted and the disturbance or noise far from the line was deleted. At last, we got the exact line parameters after fitting the points within each sets with LSM. This arithmetic performs well when an image has strong background noise. Furthermore, certain bending lines can be detected precisely. Experiments with the dynamic recognition of train brake beam show that the method is feasible and effective.