Minimizing FPS Degradation in Multi-Streaming Processing with YOLO

For the real-time object detection and tracking, CNN-based YOLO(You Only Look Once) is widely used. However, employing YOLO in a multi-streaming environment poses a challenge due to the increasing demand for computing resources as the number of streams increases. This paper proposes a system that minimizes FPS degradation without necessitating additional computing resources when using YOLO in a multi-streaming environment. Experimental results show that the proposed system can handle multi-streaming with 8 CCTV cameras without encountering FPS degradation, while maintaining delays at an acceptable level.