A Practical Approach for Vehicle Speed Estimation in Smart Cities

Abstract

The last few decades have witnessed the increasing deployment of digital technologies in the urban environment with the goal of creating improved services to citizens especially related to their safety. This motivation, enabled by the widespread evolution of cutting edge technologies within the Artificial Intelligence, Internet of Things, and Computer Vision, has led to the creation of smart cities. One example of services that different cities are trying to provide to their citizens is represented by evolved video surveillance systems that are able to identify perpetrators of unlawful acts of vandalism against public property, or any other kind of illegal behaviour. Following this direction, in this paper, we present an approach that exploits existing video surveillance systems to detect and estimate vehicle speed. The system is currently being used by a municipality of Sardinia, an Italian region. An existing system leveraging Convolutional Neural Networks has been employed to tackle object detection and tracking tasks. An extensive experimental evaluation has been carried out on the Brno dataset and against stateof-the-art competitors showing excellent results of our approach in terms of flexibility and speed detection accuracy.