基于机器视觉的钢轨表面缺陷检测系统

Design of rail surface defect detection system based on machine vision

为了实现钢轨表面缺陷自动化检测,以手推式轨道检测车为平台建立基于机器视觉的钢轨表面缺陷检测系统。以黑白线阵CCD相机获取轨道图像,并辅以线阵白光光源主动照明,减少外界光源的干扰;编码器安装在车轮上获取检测车的运动信息,并将里程转换成脉冲信号触发线阵相机进行图像采集,获取的图像经千兆网传输至工业计算机进行处理;利用单片机设计信号处理器读取编码器的脉冲信号,根据编码器A、B相输出信号的相位差判断车轮的前进方向;提出改进最大熵阈值分割法对钢轨图像分割,使分割结果接近目标出现的概率。实验结果表明,图像采集系统能够稳定的采集轨道多部件图像,相比于Otsu、原始最大熵阈值分割法,改进最大熵阈值分割法在减少误分割的同时能够比较准确的将缺陷分割出来,与背景差分法、积分投影法相比,本文方法获得了较低的误检率和漏检率。

In order to detect rail surface defects automatically,a machine vision based on rail surface defects detection system was established on the platform of hand-propelled rail vehicle.The railway images are acquired by a monochrome linear CCD camera,and the white linear light source is employed to provide illumination for the camera and reduce the influence of external light source.The encoder is installed on the wheels to obtain the motion information of the vehicle,and the mileage is converted into pulse signal to trigger the line-array camera for image acquisition.The collected images are transmitted to the computer through the gigabit network for processing.The signal processor based on the single-chip microcomputer is designed to read the pulse signal of the encoder and judge the moving direction of the wheel according to the phase difference of the output signals of A and B of the encoder.An improved maximum entropy threshold segmentation method is proposed to segment the rail image and make the segmentation result close to the probability of the occurrence of the defect.The experimental results show that the image acquisition system can acquire railway images stably.Compared with Otsu,maximum entropy threshold method,the improved method can accurately segment defects from the rail image while reducing false segmentation.Compared with other rail defect detection methods such as background difference method and integral projection method,our method obtain lower false detection rate and missed detection rate.