基于车辆跟踪的高速公路全天候违停检测算法  

Expressway All-weather Illegal Parking Detection Algorithm Based on Vehicle Tracking

作　　者：马军 刘大海 薛梦婕 李国瑞 谭圣才 MA Jun;LIU Da-hai;XUE Meng-jie;LI Guo-rui;TAN Sheng-cai(Ningxia Jiaotou Expressway Management Co.,Ltd.,Yinchuan,Ningxia 750000,China;Guangzhou Guojiao Runwan Traffic Information Co.,Ltd.,Guangzhou,Guangdong 510000,China;Highway Monitoring&Response Center,Ministry of Transport of the People’s Republic of China,Beijing 100000,China;South China University of Technology,Guangzhou,Guangdong 510641,China)

机构地区：[1]宁夏交投高速公路管理有限公司,宁夏银川750000 [2]广州国交润万交通信息有限公司,广东广州510000 [3]交通运输部路网监测与应急处置中心,北京100000 [4]华南理工大学计算机科学与工程学院,广东广州510641

出　　处：《公路交通科技》2024年第11期78-85,共8页Journal of Highway and Transportation Research and Development

摘　　要：在复杂的白天和夜晚高速公路环境下,有效地检测违停车辆尤其具有挑战性,聚焦于解决这一问题,针对高速公路监控视频中的车流量大、密度高以及光照变化等复杂因素,提出了一种适用于白天和夜晚环境的全天候车辆违停检测算法。通过构建高速公路车辆图像数据集,结合全局注意力机制和目标检测模型,并针对夜间图像添加数据增强策略,克服了在这些复杂环境下缺乏高质量车辆图像的问题,显著提高了车辆检测的检测精度。在跟踪算法上,使用匈牙利算法预测车辆位置,并根据高低分匹配框进行更新,同时使用特征匹配模型提取车辆特征,对车辆进一步跟踪,实现对疑似违停车辆的判定,最终在构建的存在违停和未存在违停的监控视频测试数据集上效果最好,算法的准确率和召回率达到了97.32%和96.57%,验证了该算法在实际应用中的优越性和可行性。基于车辆跟踪的高速公路全天候违停检测创新之处在于提出了针对高速公路监控视频的全天候违停检测算法,并在试验中取得了较好的检测精度和性能表现,在夜间能准确检测出车辆违停,并已成功应用于交通监控云平台,实现交通事故的及时预警和快速救援,有效保障交通安全和提升交通运输管理水平。Effectively detecting illegal parking vehicles is particularly challenging in complex daytime and nighttime expressway environments.Focusing on solving this problem,this study proposed an all-weather illegal parking detection algorithm for daytime and nighttime environments,targeting the complex factors,e.g.,high traffic volume,high density,and lighting variations in expressway surveillance videos.By constructing the expressway vehicle image dataset,combining the global attention mechanism and target detection model,and adding data enhancement strategies for nighttime images,the lack of high-quality vehicle images in these complex environments was overcome.The vehicle detection accuracy was significantly improved.The Hungarian algorithm was used to predict the vehicle location.The update was realized according to the high and low score matching frames.Simultaneously,the feature matching model was used to extract the vehicle features,and further track the vehicle to realize the determination on suspected illegal parking vehicle.The result indicates that it works best on the constructed test dataset of surveillance videos with and without parking violation.The algorithm accuracy and recall rate reach 97.32%and 96.57%,which verifies the superiority and feasibility of the algorithm in practical application.The innovation of all-weather illegal parking detection based on vehicle tracking is to propose an all-weather illegal parking detection algorithm for expressway surveillance video,achieving good detection accuracy and performance in the test.It can accurately detect the illegal parking at night,and has been successfully applied to the traffic monitoring cloud platform.It provides strong support for early warning and rapid rescue of traffic accidents,effectively ensuring the traffic safety and improving the traffic management level.

关 键 词：智能交通 车辆违停 全天候 车辆跟踪 夜间监控 

分 类 号：U495[交通运输工程—交通运输规划与管理]