Drone-based Computer Vision-Enabled Self-Calibrated Traffic Flow Parameter Estimation and Analysis Using Machine Learning Approaches
Date
2020
Authors
Contributor
Advisor
Department
Instructor
Depositor
Speaker
Researcher
Consultant
Interviewer
Narrator
Transcriber
Annotator
Journal Title
Journal ISSN
Volume Title
Publisher
University of Hawaii at Manoa
Volume
Number/Issue
Starting Page
Ending Page
Alternative Title
Abstract
The acquisition of traffic system information, especially the vehicle speed and trajectory information, is of great significance to the study of the characteristics and management of the traffic system. The traditional method of relying on video analysis to obtain vehicle number and trajectory information has its application scenarios, but the common video source is often a camera fixed on a roadside device. In the videos obtained in this way, vehicles are likely to block each other, which seriously affects the accuracy of vehicle detection and the estimation of speed. Although there are methods to obtain high-view road video by means of aircraft and satellites, the corresponding cost will be high. Therefore, considering that drones can obtain high-definition video at a higher viewing angle, and the cost is relatively low, we decided to use drones to obtain road videos to complete vehicle detection. In order to overcome the shortcomings of traditional object detection methods when facing a large number of targets and complex scenes, our proposed method uses convolutional neural network technology. We modified the YOLO v3 network structure and used a vehicle data set captured by drones for transfer learning, and finally trained a network that can detect and classify vehicles in videos captured by drones. At the same time, a self-calibrated road boundary extraction method based on image sequences was used to extract road boundaries and filter vehicles to improve the detection accuracy of cars on the road. Using the results of neural network detection as input, we use video-based object tracking to complete the extraction of vehicle trajectory information. Finally, the number of vehicles, speed and trajectory information of vehicles were calculated, and the average speed and density of the traffic flow were estimated on this basis. In our actual experimental results, the proposed model can achieve a detection accuracy of more than 98% under different weather, lighting conditions and traffic flow scenarios. On the basis of the obtained vehicle trajectory data, we can also complete the estimation of the average speed and average density of the traffic flow on the test road. With reference to the Level of Service (LOS) measurement index for ordinary expressways and the actual speed limit of the test road, the average density of vehicles calculated can accurately reflect the degree of vehicle congestion on the road, which can provide a reference for the intelligent traffic management system. The data obtained from the experiment can be used as a reference data set for studying the safety index of freeway traffic system and connected vehicles after manual calibration.
Description
Keywords
Traffic flow--Measurement, Motor vehicles--Speed--Measurement, Neural networks (Computer science), Drones (Aircraft)
Citation
Extent
Format
Geographic Location
Time Period
Related To
Related To (URI)
Table of Contents
Rights
Rights Holder
Local Contexts
Collections
Email libraryada-l@lists.hawaii.edu if you need this content in ADA-compliant format.