基于深度学习的高铁接触网定位器检测与识别

发布时间：2018-05-27 03:36

  本文选题：定位器 + 目标检测　； 参考：《中国科学技术大学学报》2017年04期

【摘要】：高铁接触网安全监测的主要方法是采用可见光高清相机捕捉接触网零部件的图像序列,通过图像处理和计算机视觉技术实现对零部件的检测、识别与跟踪.在整个监测系统中,定位器检测识别是必要的基础工作.传统的目标检测算法受限于特征描述子的设计,难以依靠人工设计出具有通用性、鲁棒性、高精度的特征描述子.于是提出基于Faster R-CNN模型实现高精度的接触网定位器检测,同时采用Hough变换检测出定位器的骨架轮廓,并通过滤线机制筛选出定位器的最优拟合直线段,为定位器坡度的非接触式精准测量做好基础性工作.
[Abstract]:The main method of safety monitoring for high speed catenary is to capture the image sequence of catenary parts by using visible light high-definition camera, and to detect, identify and track the parts by image processing and computer vision technology. In the whole monitoring system, locator detection and identification is a necessary basic work. The traditional target detection algorithm is limited by the design of feature descriptors, so it is difficult to design feature descriptors with universal, robust and high precision by manual design. Therefore, based on the Faster R-CNN model, the high precision detection of the catenary locator is proposed. At the same time, the skeleton profile of the locator is detected by Hough transform, and the optimal fitting line segment of the locator is selected by the filter mechanism. Do a good job for the non-contact precision measurement of the positioner slope.
【作者单位】： 中国科学院自动化研究所;中国科学院大学;
【基金】：国家自然科学基金(61432008,61532006,61472423)资助
【分类号】：TP391.41;U225
，

本文编号：1940264