Neural Network Technology for Traffic Flow Analysis in Automated Traffic Control Systems

Abstract

The article presents the results of solving the problem of developing a neural network technology for analyzing traffic flows in automated traffic control systems. Methods for analyzing traffic flows using various technologies, existing systems for analyzing traffic flows based on neural network technologies and technologies for detecting and tracking objects on a video stream in order to ensure traffic safety are described. The algorithms used in the operation of neural network technology are described, including the stages of object detection; tracking of detected objects; identifying incidents; automatic collection of information from video streams; collecting statistics. According to the results of approbation (testing) of the implemented neural network technology on own (prepared) and used video data downloaded from the Internet, the reliability of the results of the study of video frames (correct recognition of traffic flow objects) in the neural network traffic control system was 85—90 %. Errors occurred in the presence of a large number of objects on a video frames and poor quality of the video stream. When monitoring the parameters of the traffic flow, the collection, analysis and provision of aggregated data on the traffic situation on the road section were ensured. Automatic collection of information from video files is implemented up to the file time without refer­ence to real time. The statistical data obtained during the study can be used in automated traffic control systems to analyse traffic flows and in other control and decision-making systems to increase the responsiveness to incidents. The results obtained can be applied in traffic flow studies to obtain accurate statistics and forecast anomalies and incidents (traffic accidents, traffic jams) in order to ensure traffic safety.