SMARTLET: A Dynamic Architecture for Real Time Face Recognition in Smartphone Using Cloudlets and Cloud

Abstract

Abstract Face recognition in smartphone has become an important utility in a smart city for ensuring security by law enforcement. It has various applications such as capturing real life events, tracking movement of a celebrity, detecting wanted criminals, searching for missing child, surveillance, etc. Involving smartphones to do this job is challenging. Low computation power and limited battery life are the major barriers behind completing this task in real time. Various methods have already been proposed to offload computation to cloudlet or cloud to increase efficiency. However, offloading doesn't always result in performance gain. Besides, challenges still lie in different cases such as sudden disconnection from cloudlets, involvement of multiple cloudlets, load balancing between cloudlets, prefetching data to cloudlets, etc. In this paper, we have proposed a dynamic architecture named SMARTLET that distributes tasks to cloudlets based on the runtime characteristics of the communication latencies and handles aforementioned challenges efficiently. We validated our architecture by running experiments on our testbed with several smartphones, cloudlets and AWS cloud. Results show that our proposed architecture gets around 3.1× performance gain over state-of-the-art face recognition in smartphone and 21% reduction in response time with respect to the best recent architecture for offloading face recognition to the cloud.