基于光流引导Transformer模型的重载铁路监控压缩视频质量增强方法  

Compressed video quality enhancement method for heavy-haul railway surveillance based on optical flow guided Transformer model

作　　者：王文斌 宋宗莹[1] 柴雪松 凌烈鹏 李健超 WANG Wenbin;SONG Zongying;CHAI Xuesong;LING Liepeng;LI Jianchao(China Shenhua Energy Company Limited,Beijing 100080,China;Railway Engineering Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;Zhongtie Science&Technology Development Co.Ltd.,Beijing 100081,China;State Key Laboratory of High-speed Railway Track System,Beijing 100081,China)

机构地区：[1]中国神华能源股份有限公司,北京100080 [2]中国铁道科学研究院集团有限公司铁道建筑研究所,北京100081 [3]中铁科学技术开发有限公司,北京100081 [4]高速铁路轨道系统全国重点实验室,北京100081

出　　处：《铁路计算机应用》2025年第1期27-33,共7页Railway Computer Application

基　　金：中国神华科技创新基金项目(SHGF-22-4)。

摘　　要：重载铁路视频监控系统的不断扩增,使得铁路视频数据急剧增长,对数据存储和传输等能力的要求更高。为此,提出了一种基于光流引导Transformer模型的重载铁路监控压缩视频质量增强方法。通过光流补全网络提取帧间运动信息,指导Transformer模型关注视频序列中的重要特征;结合多头自注意力机制和时间空间特征融合策略,有效提取视频帧的时空特征;通过在Transformer模型结构中融入光流引导的特征增强模块,进一步提升视频质量增强的准确性和效率。基于实际采集的重载铁路监控视频数据集的实验结果表明,该方法显著优于现有的视频质量增强方法,具有实用价值。The continuous expansion of heavy-haul railway video surveillance systems has led to a sharp increase in railway video data,with higher requirements for data storage and transmission capabilities.To this end,this paper proposed a compressed video quality enhancement method for heavy-haul railway surveillance based on optical flow guided Transformer model:Extracting inter frame motion information through optical flow completion network to guide the Transformer model to focus on important features in the video sequence;Combining multi head self-attention mechanism and spatiotemporal feature fusion strategy,effectively extracting spatiotemporal features of video frames;By incorporating optical flow guided feature enhancement modules into the Transformer model structure,further improving the accuracy and efficiency of video quality enhancement.The experimental results based on the actual collected heavy-duty railway surveillance video dataset show that this method is significantly superior to existing video quality enhancement methods and has practical value.

关 键 词：重载铁路 视频增强 光流 Transformer模型 多头自注意力机制 

分 类 号：U239.4[交通运输工程—道路与铁道工程] TP391[自动化与计算机技术—计算机应用技术]