Real Time Face Recognition on an Edge Computing Device

Face recognition systems have vast applications in surveillance systems and human-computer interactions. Different approaches such as Principal Component Analysis, Fisher linear discriminant analysis, Convolutional Neural Networks (CNN) have been commonly used for face recognition. However, in the recent times, CNN's have shown quite promising results in various face recognition systems. But, deep learning based CNNs have many limitations such as they require extensive training data, have excessively high computational and cooling requirements, and lack flexibility in deployment. Fields such as robotics and embedded systems that deploy face recognition systems have significantly less power on board and limited heat dissipation capacity. Therefore, it becomes difficult to deploy deep learning models on them but edge computing based devices like the Intel Neural Stick bridge this gap as they have certain advantages. In this paper, we review different applications of face recognition systems and various algorithms used for face recognition. We then elaborate the limitations of deep learning based face recognition systems and examine how edge-computing devices can solve these problems. We then present a flowchart to deploy a CNN based face recognition model on an edge-computing device.