01657465

Component

The trajectories of all of the vehicles in congested traffic are essential in analyzing traffic dynamics. To conduct a reliable analysis with the trajectory data, we need a framework to extract those efficiently and accurately. Unfortunately, obtaining the accurate trajectories in congested traffic is a challenging problem due to false detections and tracking errors caused by adjacent vehicles. Unmanned Aerial Vehicles (UAVs), which incorporate machine learning and image processing, can mitigate these difficulties using their ability to hover above the traffic. However, there is still a lack of research on trajectory extraction in congested traffic and its evaluation. In this study, the authors propose a supervised learning-based framework for extracting vehicles’ trajectories from UAV images in congested traffic condition. To achieve this goal, they firstly develop a specialized post-processing method based on motion constraints to remove tracking errors. Then, they conduct detailed performance analyses to confirm the availability of the framework on a congested expressway in South Korea. In the best performance, the vehicle trajectories achieve as low as 0.6 m of location error with the accuracy of 86% considering misses, false-positives, and mismatches based on the training of only a small amount of positive samples. The authors' framework can be used to construct the datasets for traffic dynamics. This study contributes to the research on intelligent transportation systems as well as microscopic traffic phenomena.

This paper was sponsored by TRB committee ABJ50 Standing Committee on Information Systems and Technology. Alternate title: Automated Framework for Vehicle Trajectory Extraction in Congested Traffic Using Unmanned Aerial Vehicles

18-00932

English

8p

2018

Transportation Research Board 97th Annual Meeting

Digital/other

Figures; Photos; References; Tables

Automatic data collection systems; Automatic tracking; Automatic vehicle detection and identification systems; Data analysis; Drones; Highway traffic; Image processing; Machine learning; Microscopic traffic flow; Traffic congestion; Traffic surveillance; Vehicle trajectories

South Korea

Data and Information Technology; Highways; Operations and Traffic Management; Planning and Forecasting

Transportation Research Board Annual Meeting 2018 Paper #18-00932

TRIS, TRB, ATRI

Jan 8 2018 10:14AM