Multiple Face Detection and Recognition on Embedded Computer Vision System

Abstract

Face recognition is widely used in many applications such as biometric for authentication, surveillance system, user-identification, and personalized technology. The state-of-the-art algorithm based on Convolutional Neural Network (CNN) can achieve up to 99% of recognition accuracy. However, there is a limitation to implement the CNN based technique into embedded system to recognize multiple face in real-time as it requires extensive computation. In this thesis, we propose a framework for multiple face recognition which consists of face detection algorithm, face recognition, and tracking. Our face recognition algorithm based on state of the art deep CNN with small computational parameters. The tracking is very effective to track the face within the scene. These lead to the reduction of the recognition processing time. We implemented the proposed framework into the NVIDIA Jetson TX2 board. The experimental results show that our proposed framework can recognize multiple faces up to 8 faces at the same time in real time with 5-10 fps of processing time and the recognition rate up to 90%.