Face Detection and Face Re-identification System Using Deep Learning and OpenVINO

Face detection and face re-identification are two important processes commonly used in the surveillance system especially in Intelligent Surveillance System to identify and track people from the video footages collected from a network surveillance system that are stored inside a centralized server. This system will be helpful in crime detection, criminal tracking, and factory visitor tracking if it can be done in real time and has high accuracy. It will open the potential to revolutionize the industry and the way businesses work today. In this paper, an embedded system equipped with fast and accurate face reidentification using a modified single shot detector (SSD), a landmark regressor and a MobileNet V2 model is proposed. From the result obtained, the proposed system is tested on multiple faces and is able to achieve 99.01% re-identification accuracy. Furthermore, the proposed system also achieves an average processing time of 60ms using the low-cost Intel UP board with Intel Atom CPU and OpenVINO (a neural network acceleration platform).