Real-Time Crowd Monitoring System for Determining Level of Service Within an Area Using YOLO Algorithm

Crowd detection system is one of the more popular object detection systems. The reason for this is the danger of overcrowding that can lead to multiple deaths. To determine the level of crowd that are present in an area, a parameter called level of service is used. However, this process is often still done manually by predicting and roughly estimating the number crowd in an area. A real-time monitoring system is proposed to help automatically calculate the level of service on an area. A custom trained YOLO algorithm is used for the system to handle crowd detection. The methodology that are used in this research is software engineering. The proposed system is tested on two completely different pieces of hardware to test the system's portability and efficiency. The result is that the system manages to run on a personal computer and an NVIDIA Jetson TX2 with different efficiency. From the two hardware, the system runs much faster on personal computer with a performance of an average 80°C GPU temperature, 11.9% GPU utilization, and 45.4% memory utilization on an average of 21ms inference time.