基于相关滤波的铁路异物侵限跟踪方法研究  

Railway foreign object intrusion limit tracking method based on correlation filtering

作　　者：衣晚卓 黄克勇 王锡奎 YI Wanzhuo;HUANG Keyong;WANG Xikui(School of Communication and Signal,Nanjing Institute of Railway Technology,Nanjing 210031,China)

机构地区：[1]南京铁道职业技术学院通信信号学院,南京210031

出　　处：《铁路计算机应用》2024年第5期14-19,共6页Railway Computer Application

基　　金：南京铁道职业技术学院青年基金(Yq220008)。

摘　　要：针对铁路异物侵限频繁发生导致的列车运行安全问题,提出一种基于背景感知相关滤波器的铁路异物侵限跟踪方法。利用方向梯度直方图(HOG,Histogram of Oriented Gradient)特征提取铁路侵限异物自身特征,结合剪裁矩阵,以增加视频帧中实际背景的负样本;使用交替方向乘子法(ADMM,Alternating Direction Method of Multipliers)训练背景感知相关滤波器,减少计算复杂度,在保证跟踪速度的前提下,提升跟踪侵限异物的准确性,从而适应铁路沿线环境中由于侵限异物的形变、快速移动或天气等原因造成的目标丢失及跟踪框漂移等情况。实验结果表明,该方法对铁路侵限异物的跟踪精确度和AUC(Area Under Curve)值分别达到93%和71.9%,均高于SRDCF、KCF、ASLA和CSK等算法,具有更好的准确性。This paper proposed a railway foreign object intrusion limit tracking method based on background perception correlation filters to address the safety issues of train operation caused by frequent foreign object intrusion limit in railways.The paper utilized the HOG(Histogram of Oriented Gradient)feature extraction to extract the characteristics of railway intrusion foreign objects,and combined it with the clipping matrix to increase the negative samples of the actual background in the video frame,used the ADMM(Alternating Direction Method of Multipliers)to train background perception related filters model,reduce computational complexity and improving the accuracy of tracking intruded foreign objects while ensuring tracking speed,thus adapted to the situations of target loss and tracking frame drift caused by deformation,rapid movement or weather of intruded foreign objects in railway environments.The experimental results show that the tracking accuracy and AUC(Area Under Curve)value of this method for railway intrusion foreign objects reach 93%and 71.9%,respectively,which are higher than algorithms such as SRDCF,KCF,ASLA,and CSK,and have better accuracy.

关 键 词：铁路异物侵限 跟踪方法 相关滤波 方向梯度直方图(HOG)特征 剪裁矩阵 

分 类 号：U216.3[交通运输工程—道路与铁道工程] TP399[自动化与计算机技术—计算机应用技术]