An Accurate and Efficient Crowd Detection Model for Smart Energy Management Based on YOLOv5

The energy control of classrooms on campus plays a key role in reducing the energy cost and improving university management. However, the complexity of classroom environments leads to several issues such as low accuracy of crowd detection, which impedes the identification of status in the classroom. As a result, it is difficult for administrators to determine the usage of classrooms, which may cause a huge amount of energy and time waste. In this paper, we propose a model to promote the accuracy and efficiency of crowd detection in the classroom scenario by applying the YOLOv5 algorithm. Specifically, we first collect videos recorded in the classrooms from different periods. Then we generate image samples with the same frame rate as the collected videos. The model is finally produced based on the YOLOv5 algorithm. Results show that the trained model achieves an accuracy of 94.3% with little overhead, indicating that the model is effective and accurate in the classroom scenario. We also demonstrate a practical use case in which the proposed model can help the energy control system to adjust the usage of electrical appliances (e.g., air conditioners and lights) in the classroom, which further contributes to the development of low-carbon living and green energy-saving campus.