Hierarchical Convolutional Neural Network Architecture in Distributed Facial Recognition System

Authors propose an efficient distributed facial recognition system based on the cascade of Convolutional Neural Networks. The traditional approach utilizing the centralized schema of recognition system has its main disadvantage concluding a high traffic and computing load of the central processing server: the more video surveillance cameras are served by the central unit, the more its load rate. But most of the time the high-quality video stream does not consist of facial information and the computational resources are being wasted. The main principle of our recognition system is the distributed hierarchical processing network utilizing the "coarse-fine" paradigm. Each source video stream is processed in-place by the tiny SoC computer which acts as an Edge Computing Unit and detects the presence of a face fragment in a video frame and crops the bounds of the ROI. The resulting stream including ROI is relayed to the main server if the face is detected.