Abstract: The use of the Internet of Things (IoT) is steadily increasing in a wide range of applications. Among these applications, safety and security are some of the prominent applications. Through surveillance systems, we can restrict access to our premises and thus secure our assets. Nowadays face detection and recognition enabled surveillance systems are available in the market, which can detect faces from video frames captured using IP cameras, and then recognize those faces by comparing them with existing databases. However, higher prices and low accuracy are impeding the large scale deployment of those systems. In this paper, we have proposed a generic architecture for face detection and recognition system from real-time video frames that have been captured through IP cameras and processed using low-cost IoT devices by utilizing Cloud computing services. We have selected two IoT platforms: Eclipse Mosquitto IoT broker and Kaa IoT middleware to implement our proposed architecture. The face detection part is deployed in the IoT devices and the computation-intensive task, i.e., face recognition is carried out in backend Cloud servers. We have executed our experiments in two different Cloud infrastructures: Core Cloud and Edge Cloud and measured the total processing time in different scenarios. The experimental results show that the performance of the Mosquitto broker in terms of total processing time is better than Kaa middleware. Total processing time can be further reduced by deploying a face recognition application from Core Cloud to Edge Cloud. Furthermore, the k-nearest neighbor algorithm shows promising results compared to other face recognition algorithms.
Keywords: IoT platforms, real-time, face recognition, Cloud computing, computation offloading, response time
