Detection of traffic anomalies for a safety system of smart city

Abstract

For modern smart city with sustainable development we need to provide reasonable level of safety and efficient management of the resources. Instant response to incidents and abnormal situations will help to provide such high bars for city residents, which requires deployment of application of intelligent information processing and data analytics into infrastructure. Closed-circuit television (CCTV) is playing a key part in assurance of city security - most of the modern large cities equip with powerful monitoring systems and surveillance cameras. Video data covers most of the city and could be efficiently used to find anomalies or trends. This hard task for non-stop video monitoring could be solved by modern achievements in machine learning and computer vision techniques, which can automate the process of video analysis and identify anomalies and incidents without human intervention. In this paper, we used computer vision methods like object detection and tracking, as well as neuron networks for classification and detection of anomalies on real time video. As a result of this work we suggested the working approach for detection of vehicle/pedestrian violating legal trajectory anomaly, which we tested on real-time video provided by surveillance cameras of the city of Kazan.