摘要
基于计算机视觉的车载轨道巡检系统由图像数据采集、数据分析和信息管理3个子系统组成;采用线阵相机进行1.6mm等间距运动扫描,并运用多线程交互和虚拟内存映射技术实现运动状态下轨道图像数据的采集和存储;在分析轨道图像病害特征的基础上,运用主成分分析、线性判别分析和Adaboost等方法建立机器学习模型,实现对钢轨、扣件和道床3个区域病害的模式识别。应用验证表明:系统能够有效检出钢轨表面擦伤、扣件缺失和移位以及轨道板裂纹等病害,其中钢轨表面擦伤、扣件缺失的检出率达95%;能够代替传统的人工步行巡道,而且最高巡检速度可达160km.h-1。
The on-board track inspection system based on computer vision is composed of image data acqui- sition, data analysis and information management. A movably mounted linear array camera is used to mo- tion scan the track at the equal interval of 1.6 mm. Combining the multi-thread interaction and virtual memory mapping technology, the image data of track can be acquired and stored under motion state. Ac- cording to the analysis of the disease characteristics in track image, the machine learning model is estab- lished by using principal component analysis, linear discriminant analysis and Adaboost algorithms. Then the pattern recognition of the defect of rail, fastening and ballast bed is implemented. The application dem- onstrates that the system can detect a number of defects, such as rail surface defect, fastening missing or shifting, track slab crack and so on. The detection rate of rail surface defect and fastening missing is above 95 %. The system can replace the traditional man-walking inspection, and the highest speed of inspection can reach 160 km · h-1.
出处
《中国铁道科学》
EI
CAS
CSCD
北大核心
2013年第1期139-144,共6页
China Railway Science
基金
国家"八六三"计划项目(2011AA11A102)
关键词
轨道巡检
计算机视觉
线阵相机
高速铁路
Track inspection
Computer vision
Linear array camera
High-speed railway
