| 基于地空信息融合的无人机交通状态感知方法研究 |
| |
| 引用本文： | 黄玲,吴泽荣,洪佩鑫,张荣辉,吴建平.基于地空信息融合的无人机交通状态感知方法研究[J].中国公路学报,2021,34(12):249-261. |
| |
| 作者姓名： | 黄玲 吴泽荣 洪佩鑫 张荣辉 吴建平 |
| |
| 作者单位： | 1. 华南理工大学 土木与交通学院, 广东 广州 510640;2. 东南大学 现代城市交通技术江苏高校协同创新中心, 江苏 南京 210096;3. 中山大学 广东省智能交通系统重点实验室, 广东 广州 510275;4. 清华大学 土木工程系, 北京 100084 |
| |
| 基金项目： | 广东省区域联合基金重点项目（2020B1515120095）；广州市重点领域研发计划项目（202007050004）；广东省普通高校特色创新类项目（2019KTSCX007）；广东省自然科学基金项目（2021A1515011794）；国家自然科学基金项目（52172350，51408237，51775565） |
| |
| 摘 要： | 现有的无人机（UAV）交通状态感知方法，主要针对宏观交通状态参数的获取，同时尚未克服UAV自运动对交通参数检测精度的影响，难以满足智能交通系统对于高精度微观交通参数的应用需求。为此，提出一种基于地空信息融合的UAV交通状态感知方法，该方法包括：地空信息融合模型、道路关键点（IKP）检测及跟踪、车辆目标检测及追踪算法和交通状态参数提取及估计。其中，地空信息融合模型利用地基信息（IKP世界坐标）与空基信息（IKP像素坐标）进行最优化融合，并通过自适应IKP追踪算法与自适应UAV位置偏移判断算法实时更新模型参数，以此克服UAV自运动对车辆轨迹精度的影响，进而获取可靠的车辆级（瞬时速度、车头间距和车头时距）与车道级（车道动态密度、车道流量和空间平均车速）交通状态参数。利用提出的感知方法获取实地拍摄视频的车辆级交通参数并进行了分布检验，同时比较了基于不同交通流模型的车道级参数估算方法。结果表明：该方法在车辆检测的mAP@0.5指数超过90%，同时提取的车辆轨迹相对完整，获取的车辆级和车道级交通状态参数也符合实际交通流状况。最后，将该模型应用于实地道路的交通拥堵检测及交通事件检测，该研究结果为UAV在现代交通感知和管理中的应用提供了一种理论和技术参考。
|
| 关 键 词： | 交通工程 无人机交通感知 地空信息融合 交通状态参数 车辆目标检测 深度学习 |
| 收稿时间： | 2021-04-29 |
Research on UAV Traffic State Perception Method Based on Air-ground Information Fusion |
| |
| HUANG Ling,WU Ze-rong,HONG Pei-xin,ZHANG Rong-hui,WU Jian-ping.Research on UAV Traffic State Perception Method Based on Air-ground Information Fusion[J].China Journal of Highway and Transport,2021,34(12):249-261. |
| |
| Authors: | HUANG Ling WU Ze-rong HONG Pei-xin ZHANG Rong-hui WU Jian-ping |
| |
| Affiliation: | 1. School of Civil Engineering and Transportation, South China University of Technology, Guangzhou 510640, Guangdong, China;2. Modern Urban Transportation Technology Jiangsu University Collaborative Innovation Center, Southeast University, Nanjing 210096, Jiangsu, China;3. Guangdong Key Laboratory of Intelligent Transportation System, Sun Yat-sen University, Guangzhou 510275, Guangdong, China;4. Department of Civil Engineering, Tsinghua University, Beijing 100084, China |
| |
| Abstract: | The existing UAV (Unmanned Aerial Vehicle) traffic state perception method is mainly aimed at the acquisition of macroscopic traffic state parameters and has not overcome the impact of UAV self-motion on the detection accuracy of traffic parameters. It is difficult to meet the application requirements of intelligent transportation systems for high-precision microscopic traffic parameters. To this end, a UAV traffic state perception method based on air-ground information fusion is proposed. The method includes air-ground information fusion model, infrastructure key points (IKPs) detection and tracking, vehicle detection and tracking algorithm, and traffic state parameter extraction and estimation. The air-ground information fusion model uses ground-based information (IKP world coordinates) and air-based information (IKP pixel coordinates) for optimal fusion and updates the real-time parameters through the adaptive IKP tracking and offset estimating algorithm. This overcomes the influence of UAV self-motion on the accuracy of vehicle trajectory and obtains reliable vehicle-level (instantaneous speed, headway, and headway time) and lane-level (lane dynamic density, lane flow, and space average vehicle speed) traffic state parameters. The proposed perception method is used to obtain and validate the distribution of the vehicle-level traffic parameters of the field video. At the same time, the estimation methods of lane-level parameters based on different traffic flow models are compared. The results show that the mAP@0.5 index of vehicle detection exceeds 90%, the extracted vehicle trajectories are relatively intact, and the obtained vehicle-level and lane-level traffic state parameters conform with the actual traffic flow conditions. Finally, the model is applied to the traffic jam and incident detection. The research results provide theoretical and technical support for the application of UAV in modern traffic perception and management. |
| |
| Keywords: | traffic engineering UAV traffic state perception air-ground information fusion traffic state parameter vehicle detection deep learning |
|
| 点击此处可从《中国公路学报》浏览原始摘要信息 |
|
点击此处可从《中国公路学报》下载全文 |
|
