Dynamic Tasks Assignment for Face Recognition in Edge Computing

Abstract

Face recognition has been widely used in airport security, face payment, access control, and other scenarios. Face recognition contains a sequence of tasks, e.g. face detection, preprocessing, feature extraction and classification recognition. This paper focuses on the dynamic tasks assignment for face recognition in edge computing architecture. Our preliminary experimental results show that different ways of assigning tasks can significantly affect the response time of face recognition services. Motivated by this finding, we propose a new framework for face recognition in edge computing architecture which supports the face recognition tasks assignment to Edge server and Cloud server. We further propose a dynamic tasks allocation algorithm based on decision tree that minimizes the overall response time for face recognition. Simulation results indicate that the average response time of our scheme is 62.5%, 34.7%, and 17.4% lower than that of a simple model, cloud model, and fog computing (FC) model respectively.