Decentralized Vehicular Identification and Tracking on Lightweight IoT Edge Nodes

Video Surveillance Systems (VSSs) are among the most investigated and widely adopted systems in smart cities by administrative officials for public safety and by private sector individuals to secure residents, employees, and properties. While many new technologies based on Machine Learning (ML) have enabled systems to be utilized for a diverse set of applications, such as license plate detection and car YMM (Year, Make Model) classification, one of the underlying concerns behind VSSs for both public and private sectors is the cost and resource constraints of the devices themselves. Given that optical character recognition and object detection are already computing and data-intensive activities, both in regards to the training of those ML models, as well as the real-time processing of the data, it is highly desired to create or optimize systems to operate on lightweight Internet of Things (IoT) devices. In this paper, we introduce a lightweight scheme for Decentralized Vehicular Identification (DEVID), which is affordable to inexpensive IoT devices as long as they are capable of managing their inputs and outputs (IO) in a fashion that is not costly in terms of storage or power consumption. While the performance in the experimental study is not satisfactory, the DEVID system possesses the potential to be a cost-friendly lightweight solution for smart cities.