摘要
针对复杂环境下的钢轨伸缩调节器伸缩量测量问题,提出一种基于深度学习和视觉定位的非接触式钢轨伸缩调节器伸缩量实时测量方法。训练Faster R-CNN模型检测粘贴在伸缩调节器可活动钢枕和轨道固定轨枕上的标识牌,然后使用HSV空间的阈值分割算法和包括主成分分析法(PCA,principal component analysis)在内的筛选算法获得标识牌的圆心关键点坐标,最后通过透视变换得到伸缩量的测量结果。结果表明,该算法能在复杂环境下对钢轨伸缩调节器的伸缩量进行实时测量,拥有较高的精度和较强的鲁棒性。
Aiming at the problem of measuring the expansion amount of rail expansion regulator in complex environment,this paper proposes a real-time and non-contact measurement method of the expansion amount of rail expansion regulator based on deep learning and visual positioning.The Faster R-CNN model is trained to detect the signboard pasted on the movable steel sleeper of the rail expansion regulator and the rail fixed sleepers, and then the key point coordinates of the circle center in the signboard are obtained by using the threshold segmentation algorithm of HSV space and the filtering algorithm including principal component analysis. Finally,the measurement results of the expansion amount are obtained through perspective transformation.
出处
《工业控制计算机》
2022年第5期64-66,69,共4页
Industrial Control Computer
基金
中国教育部联合基金(6141A02022362)
湖北省技术创新重大专项(2019AAA059)
中铁第四勘察设计院集团有限公司科研课题(2020K026)。
关键词
机器视觉
钢轨伸缩调节器
Faster
R-CNN
阈值分割
主成分分析
透视变换
machine vision
rail expansion regulator
Faster R-CNN
threshold segmentation
principal component analysis
perspective transformation