Modeling and Performance Evaluation of Collaborative IoT Cross-Camera Video Analytics

Internet of things (IoT)-based cameras have been proposed for surveillance and mixed reality applications in different domains. IoT cameras stream video frames produced from the scenery within their coverage area. Video frames are then processed to determine, identify, classify, and track objects of interest. Traditionally, IoT video frames are offloaded to distant cloud servers and processed by computation-intensive algorithms for video analytics. However, this approach incurs increased latency and network overhead. Recent studies proposed edge computing for resource provisioning for IoT applications. Nevertheless, edge computing presents many daunting challenges, such as the efficient management and allocation of edge server resources to IoT cameras. We propose the collaborative work of IoT cameras for video analytics. Accordingly, idle resources on an IoT camera can be used to process video frames produced by neighboring cameras. In this regard, we devised a mathematical framework for collaborative cross-camera IoT video analytics, which allows engineers and practitioners to obtain useful insides for the further design of architectures for collaborative IoT video analytics. We conduct extensive numerical evaluations and the obtained results show that the resources' utilization, offloading latency, and offloading costs are sensitive to the neighborhood resources demand.