一种基于时空相关分析的货运列车车号识别方法研究  

Research on identification method of freight train number based on spatio-temporal correlation analysis

作　　者：王保宪 王凯 杨宇飞 李义强 赵维刚 WANG Baoxian;WANG Kai;YANG Yufei;LI Yiqiang;ZHAO Weigang(Structure Health Monitoring and Control Institute,Shijiazhuang Tiedao University,Shijiazhuang 050043,China;School of Electrical and Electronic Engineering,Shijiazhuang Tiedao University,Shijiazhuang 050043,China;Key Laboratory for Health Monitoring and Control of Large Structures of Hebei Province,Shijiazhuang 050043,China)

机构地区：[1]石家庄铁道大学大型结构健康诊断与控制研究所,河北石家庄050043 [2]石家庄铁道大学电气与电子工程学院,河北石家庄050043 [3]河北省大型结构健康诊断与控制重点实验室,河北石家庄050043

出　　处：《铁道科学与工程学报》2021年第4期999-1008,共10页Journal of Railway Science and Engineering

基　　金：河北省重点研发计划项目(19210804D);国家自然科学基金资助项目(51808358);河北省高等学校科学技术研究项目(BJ2020057);国家能源投资集团有限责任公司科技创新项目(SHGF-15-41);石家庄铁道大学研究生创新资助项目(YC2020067);石家庄铁道大学优秀青年科学基金资助项目。

摘　　要：提出一种基于时空相关分析的货运列车车号识别方法,该方法包括车号定位、片段聚类与车号识别3部分。基于连通体分析技术,提出利用货运列车车号字符间特定的几何比例关系有效地定位车号区域;在车号定位基础上,利用视频序列时空冗余信息建立帧信息补正模型,对部分定位错误帧图像进行补正并通过片段聚类方法将包含相同内容的车号视频序列进行切分;利用概率神经网络训练车号联合识别决策器,对可能包含同一车号的多帧图像进行联合识别,有效提高车号识别的准确率。通过在实际货运列车视频数据集上进行测试验证,本文算法对所有帧图像的平均车号识别准确率高于90%,优于传统基于静态图像处理的车号识别方法。In this paper,an effective and efficient freight train number identification model was processed based upon the spatiotemporal correlation analysis,which consists of three parts:train number location,fragment frame clustering and train number identification.Firstly,via the connected component analysis,the specific geometric proportion relation between these numbers characters of freight train was used to locate the train number area effectively.Secondly,on the basis of train number area location,one frame information complement framework was established by using the temporal and spatial redundancy information of video.Within this framework,the frame containing the error location result can be corrected,and the train sequences that contain the same contents were clustered by fragment clustering method.Thirdly,the probabilistic neural network was utilized for train number recognition,which jointly identifies multiple image frames which may contain the same train number,thereby improving the accuracy of train number recognition.Simulation results on the practical train video dataset demonstrate that the average recognition accuracy of our presented model is higher than 90%,which is better than the traditional static freight train number identification methods.

关 键 词：货运列车 车号定位 车号识别 时空相关性 概率神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]