Automating Road Traffic Monitoring Using Computer Vision

I. A. Chebykin

DOI: https://doi.org/10.30932/1992-3252-2020-18-6-74-87

2021-07-30

World of Transport and Transportation

Abstract:The objective of the article is to describe application of computer vision and artificial intelligence technologies for solving the problems of road infrastructure design. The article evaluates the traditional methods of quantitative and qualitative analysis of traffic flows in terms of labour intensity and accuracy using the method of comparative analysis, the advantages and disadvantages of the considered methods are indicated. A new method of traffic flow analysis using unmanned aerial vehicles and computer vision technology based on convolutional neural networks is proposed. The considered method makes it possible to fully automate collection and analysis of data on traffic flows. The article describes the first application of the proposed method when performing transport and economic surveys within the framework of the design of «Northern bypass of the city of Perm». The advantages of the applied method in relation to the traditional ones are described. To implement this project, software was developed for analysing traffic flows using video materials. Further, traffic monitoring is considered, its goals and objectives are described, the necessary functionality of the road traffic monitoring automation system is indicated, the traffic parameters that it should determine are listed. The methodology for implementation of an automated traffic monitoring system based on video materials on a section of the road is considered. A presented project of a traffic monitoring system makes it possible to extend the previously considered approach to the entire road network. Technologies are described that make it possible to implement this system based on video analytics of materials from CCTV cameras. A method for vehicle re-identification is proposed, and the implementation of this method is demonstrated. The method allows building a correspondence matrix of vehicles recorded by CCTV cameras located on different segments of the road network, as well as determining all traffic parameters for the entire street and road network. The conclusions outline the prospects for development of the developed software in terms of application in intelligent transport systems.