Path Finding Camera Network Using Edge Device

Object tracking through camera surveillance can be used as a tool for reducing criminal activities and increasing the security especially in modern city life. A prototype for a multi-camera surveillance system that can identify objects and track their movements across multiple cameras is proposed. The system was built on a platform that included a Raspberry Pi, an Edge TPU, and three cameras and has been able to accurately detect objects in a variety of conditions and scenarios. Real time comparison of two object detection algorithms, ‘you only look once’ (YOLO) and omni-scale network (OSNet) is carried out. Re-identification of the object was demonstrated with the optimal use of available memory of the system. The proposed system demonstrated high performance in terms of throughput, latency, and model accuracy, making it a viable solution for real-time multi- object tracking and re-identification without utilizing a cloud server.