基于图像检测的轨距检测算法研究

Track Detection Algorithm Based on Image Detection Research

为实现实时在线轨距测量,建立车载轨距机器视觉检测系统,对该系统所采用的去噪前置处理、精确阈值分割处理和距离变换等算法进行研究.介绍系统的工作原理与构成,并在分析传统光带中心线提取算法的基础上提出基于PNDT的快速轨道轮廓中心线提取方法;采用强对比度拉伸和指数变换的方法进行图像增强,并结合高斯平滑与动态ROI对图像进行快速去噪前置处理;对图像进行精确阈值分割处理,采用距离变换的方法得到轨道轮廓中心线并准确定位轨距测量点.实验结果表明:该系统检测精度满足-1mm^+1mm,实验室测试的合成不确定度最大为0.52mm,图像帧处理时间为14.35ms.轨距机器视觉检测系统可满足实时在线轨距检测对系统检测速度和精度的要求.

In order to achieve real-time online gauge measurement,an on-board machine vision system of gauge detection is established and its pre-noising processing,precise threshold segmentation processing and distance transform algorithms are studied.The system working principle and structure are introduced,and the PNDT fast rail contour centerline extraction method is proposed based on traditional light stripe centerline extraction algorithm.The strong contrast stretching and the exponential transform are used to enhance the image and the Gaussian smoothing and dynamic ROI are used for fast pre-noising processing.Based on the precise threshold segmentation processing of image,the rail contour centerline is obtained and the gauge measurement point is accurately located by using distance transform.Experimental results show that the detection accuracy of the system meet-1mm~＋1mm,the combined uncertainty of laboratory testing up to 0.52 mm,image frame processing time is 14.35 ms.The system can meet the requirement of detection speed and accuracy in gauge real-time online detection.