摘要
实时、精确地确定列车在轨道路径上的位置是保障行驶安全、提升运输效率、提供最佳服务的前提。为了解决传统绝对定位技术存在的一些不足,提出一种基于改进YOLOv3的轨道定位点检测方法。根据定位点目标大小,调整网络输入尺寸及其特征提取网络Darknet-53的结构;由于定位点样本数量稀缺,故采用旋转、增噪等手段进行样本扩充,并使用K-means算法对自制的训练集聚类分析;依据官方网络参数说明及实际图片特征,优化网络的训练参数。实验结果表明,改进后YOLOv3算法的召回率提高了23百分点,检测速度提升了31帧/s,即在提高定位可靠性的同时,也满足轨道巡检系统的实时检测速度要求。
The real-time and accurate determination of the position of the train on the track path is a premise for ensuring driving safety,improving transportation efficiency and providing the best services.In order to solve some shortcomings of traditional absolute positioning technology,this paper proposed a track locating point detection method based on the improved YOLOv3.According to the target size of the locating points,the input size of the network and the structure of the feature extraction network Darknet-53 were adjusted.Due to the scarce quantity of locating point samples,this paper used such methods as rotation and noise increase to expand samples,and adopted the K-means algorithm to do cluster analysis of the homemade training set;the training parameters of the network were optimized on the basis of descriptions of official network parameters and the actual image characteristics.The experimental results show that the recall rate of the improved YOLOv3 algorithm is increased by 23 percentage points,and its detection speed is increased by 31 fps,which not only improves the positioning reliability,but also meets the real-time detection requirements of the track inspection system.
作者
郑英杰
李舒婷
吴松荣
刘东
韦若禹
Zheng Yingjie;Li Shuting;Wu Songrong;Liu Dong;Wei Ruoyu(Key Laboratory of the Ministry of Education for Maglev Technology and Trains,Chengdu 610031,Sichuan,China;School of Electrical Engineering,Southwest Jiaotong University,Chengdu 611756,Sichuan,China)
出处
《计算机应用与软件》
北大核心
2021年第9期188-192,213,共6页
Computer Applications and Software
基金
四川省科技计划项目(2019JDTD0003)。
