IoT video analytics for surveillance-based systems in smart cities

Smart city applications are revolutionizing the way people interact with diverse systems in city-wide applications. Internet of Things (IoT) and machine learning are two enabling technologies for smart cities. IoT sensors are deployed for data collection from city environments and machine learning models are used for data processing, inference, and predictions aimed at controlling IoT actuators and providing information for users. Surveillance applications in smart cities often use cloud and edge servers for the data analytics of video streamed by IoT cameras. This approach might increase the system's latency due to the communication delay and networking congestion in the links connecting IoT cameras to distant cloud servers. A promising approach to mitigate the latency in video analytics applications is to perform the video analytics locally at the IoT devices and IoT gateways by using lightweight convolutional neural network (CNN) models. In this paper, we propose an architecture for implementing at IoT devices the data pipeline required for local video analytics in smart cities. Moreover, we instantiate the proposed framework to conduct an extensive performance evaluation of five CNN models on IoT devices, in a traffic surveillance smart city application. Obtained results show the feasibility of studied CNN models, in terms of frame processing rate and energy efficiency in the considered IoT devices.