Development of machine-vision technology for inspection of railroad track

machine vision;computer vision;railroad;track;inspection;Association of American Railroads (AAR);defect;ballast;fastener;anchor;turnout;University of Illinois;UIUC;Beckman Institute;Computer Vision and Robotics Laboratory (CVRL)

Sawadisavi, Steven V. ; Barkan ; Christopher P.L. ; Edwards ; John R.

关键词: machine vision;    computer vision;    railroad;    track;    inspection;    Association of American Railroads (AAR);    defect;    ballast;    fastener;    anchor;    turnout;    University of Illinois;    UIUC;    Beckman Institute;    Computer Vision and Robotics Laboratory (CVRL);   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/18420/Sawadisavi_Steven.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Railroad engineering practices and Federal Railroad Administration (FRA) regulations require track to be inspected for physical defects at specified intervals, which may be as often as thrice per week.These inspections are conducted visually by railroad track inspectors, but due to practical considerations, only a certain level of detail and consistency can be obtained.Enhancements are possible using machine-vision technology, which consists of recording digital images of track elements and analyzing those images using custom algorithms to identify defects or their symptoms.Based on analysis of FRA accident data, discussion with railroad track engineering experts, consultation with Association of American Railroads researchers, and review of existing inspection technologies and methods, this project focuses on developing a machine-vision-based system to detect irregularities and defects in wood-tie fasteners, rail anchors, crib ballast, and turnout components.A Video Track Cart was developed for initial video data acquisition, and algorithms were developed to consistently detect the rail, tie plates, ties, cut spikes, rail anchors, and ballast using a global-to-local algorithmic approach.Using the detection algorithms on panoramas generated from the videos further increases their accuracy, with added benefit in using the panoramas to manually confirm the severity of defects if results are in doubt. Once defects have been detected and catalogued by the system, a quantitative comparison of data from different runs is possible, opening up possibilities for defect growth trending and predictive maintenance scheduling.Ultimately, this system will provide consistent, quantitative track inspection data for not only increasing current inspection capabilities, but also deepening the understanding of track health over time.