基于无人机视频的交叉口流量检测方法  

Intersection traffic detection method based on UAV videos

作　　者：汤睿尧 胡春钢 欧婉情 TANG Rui-yao;HU Chun-gang;OU Wan-qing

机构地区：[1]南京理工大学自动化学院,江苏南京210094

出　　处：《智能城市》2023年第3期114-118,共5页Intelligent City

基　　金：国家重点研发计划政府间国际科技创新合作重点专项项目“基于智能网联的城市电动公交管控关键技术研究与应用”(项目编号:2019YFE0123800);中央高校基本科研业务费专项资金(项目编号:30920021142)。

摘　　要：交叉口的运行效率直接影响城市路网的运行状态,近年来随着无人机技术的兴起,其应用给交叉口运行效率的研究带来了新途径,无人机以其特有的优势为交通参数获取带来了便利。文章基于深度学习的方法采用改进的YOLOv5l+DeepSort的多目标检测追踪算法实现交叉口机动车辆的检测追踪,获取其运行轨迹及运动参数,检测精度达到0.926。在此基础上融合目标空间坐标点位移和角度变换实现对各进口道机动车不同流向流量的准确检测。最后选取南京市星火路-学府路交叉口进行验证,本方法的流量检测准确率达到95%。结果证明本方法可以对交叉口运行状态合理量化,为交通调查分析和改造优化提供技术支持。The operation efficiency of intersections is directly related to the operation status of urban road network.In recent years,with the rise of UAV technology,its application has brought a new way to the research of intersection operation efficiency.UAV brings convenience to the acquisition of traffic parameters with its unique advantages.Based on the method of deep learning,the paper uses the improved YOLOv5l+DeepSort multi-target detection and tracking algorithm to realize the detection and tracking of motor vehicles at intersections,and obtains their running tracks and motion parameters.The detection accuracy reaches 0.926.On this basis,the target space coordinate point displacement and angle transformation are fused to realize the accurate detection of the different flow directions of motor vehicles at each entrance.Finally,two intersections in Nanjing were selected for verification,and the accuracy of flow detection of this method reached 95%.The results show that this method can reasonably quantify the operation status of intersections,and provide technical support for traffic investigation and analysis and reconstruction optimization.

关 键 词：机器视觉 交通流量检测 无人机视频 多目标检测追踪 

分 类 号：U495[交通运输工程—交通运输规划与管理]