Obtaining a high-quality traffic state over the wide urban area is critical for implementing effective urban traffic control. Among the various traffic state variables, traffic density is one of the essential variables since it plays a vital role in predicting traffic congestion. With the development of edge-computing devices and computer vision technology, the vehicle-mounted camera is becoming a promising option for estimating traffic density over a vast urban area. The previous literature uses object tracking or optical flow to estimate the traffic state from the vehicle-mounted camera. However, these algorithms are hard to implement in real time due to a high computational cost and the requirement of consecutive frames for the algorithm to work. To this end, we propose a methodology that only leverages an object detection algorithm to estimate the traffic state for the real-time application. It can work on a sample of full frames instead of consecutive frames. We define the region of interest as the vehicle driving area between the detected lanes to estimate the traffic density and congestion of the corresponding region. In case the vehicle driving lane is the bus-only lane, we estimate the traffic density of the adjacent left lane. We show the effectiveness of our algorithm in urban traffic state estimation by over 90% density estimation accuracy on videos collected from two bus routes and more frequent traffic congestion in a grid-like road structure on the congestion map derived by our algorithm.
Urban Traffic Density Estimation from Vehicle-mounted Camera for Real-time Application
2023-02-20
4677391 byte
Conference paper
Electronic Resource
English
Vehicle-mounted road gradient real-time estimation method
European Patent Office | 2020
|European Patent Office | 2019
|VEHICLE-MOUNTED CAMERA AND VEHICLE-MOUNTED CAMERA SYSTEM
European Patent Office | 2016
|