针对雾霾天气的铁路调度视频图像智能化分析预警技术研究  

Research on the Intelligent Analysis and Early Warning Technology of Railway Dispatching Video Image for Hazy Weather Conditions

作　　者：周惟祥 满恒懋 段卿培 姚树金 赵宁 刘仁伟 ZHOU Weixiang;MAN Hengmao;DUAN Qingpei;YAO Shujin;ZHAO Ning;LIU Renwei

机构地区：[1]大连理工大学数学科学学院,辽宁大连116024 [2]中国铁路北京局集团有限公司调度所,北京100086

出　　处：《铁道通信信号》2024年第10期13-18,共6页Railway Signalling & Communication

基　　金：中国铁路北京局集团有限公司科技项目(2023BY01)。

摘　　要：为实现在雾霾天气下铁路调度视频图像的智能化分析预警,针对性地提出一种铁路视频异物检测方法。首先针对部分视频中存在的雾霾退化问题,引入基于深度网络的恢复模型,可得到清晰视频,用于进一步检测;其次引入YOLOv5目标检测模型,针对铁路场景进行重新训练,并采用冻结训练和串联模型策略,使模型在具有异物检测能力的情况下,保留对常规目标的检测能力;最后针对铁路视频真实数据中缺乏异物的问题,采用数据增强技术,加强铁路视频识别异物的能力。试验结果表明:该检测方法显著提升了在雾霾条件下的铁路视频异物检测效果,有助于提高人员闯入、异物入侵等安全隐患的实时识别能力。To enable intelligent analysis and early warning of railway dispatching video images in hazy weather conditions,a railway video foreign object detection method specifically designed for such environments is proposed.Firstly,a deep network-based restoration model is introduced to mitigate haze degradation in certain videos and produce clear content for subsequent detection.Secondly,the YOLOv5 target detection model is introduced and retrained for railway scenes,employing strategies of freeze training and series model to maintain detection capabilities for common targets while enhancing foreign object detection.Finally,the data enhancement technology is applied to address insufficient foreign object detection in railway video data,improving the effectiveness of railway video foreign object detection.Experimental results show that this approach significantly enhances the effectiveness of railway video foreign object detection in hazy weather conditions,thereby improving the real-time identification of security risks such as personnel intrusion and foreign object invasion.

关 键 词：铁路调度监控视频 异物检测 雾霾天气 目标检测 图像恢复 数据增强 

分 类 号：U283.2[交通运输工程—交通信息工程及控制]