ScaledYOLOv4 Based Person Re-identification System

Despite the fact that facial verification and tracking are a solved problem that is used in many fields and has many product solutions, it is not easy to find an open-source implementation of a human face tracking system with its description. In this paper, we introduce a person tracking system, describe its architecture and functionality. The proposed system is modular and scalable. It includes an IP-camera data stream reader module, face detection module based on state-of-the-art ScaledYOLOv4 model, face identification module with Dlib, and database reader-writer module for storing and autoscaling embeddings and features. The proposed system is easily integrated into existing software and hardware and easy to implement and repeat. Moreover, due to its modularity, it is easy to keep up with the state-of-the-art solutions in the face detection and identification tasks and, if necessary, implement them in the system.