Detection of Room Occupancy in Smart Buildings

Recent advancements in occupancy and indoor environmental monitoring have encouraged the development of innovative solutions. This paper presents a~novel approach to room occupancy detection using Wi-Fi probe requests and machine learning techniques. We propose a~methodology that splits occupancy detection into two distinct subtasks: personnel presence detection, where the model predicts whether someone is present in the room, and occupancy level detection, which estimates the number of occupants on a~six-level scale (ranging from 1 person to up to 25 people) based on probe requests. To achieve this, we evaluated three types of neural networks: CNN (Convolutional Neural Network), LSTM (Long Short-Term Memory), and GRU (Gated Recurrent Unit). Our experimental results show that the GRU model exhibits superior performance in both tasks. For personnel presence detection, the GRU model achieves an~accuracy of 91.8\%, outperforming the CNN and LSTM models with accuracies of 88.7\% and 63.8\%, respectively. This demonstrates the effectiveness of GRU in discerning room occupancy. Furthermore, for occupancy level detection, the GRU model achieves an~accuracy of~75.1\%, surpassing the CNN and LSTM models with accuracies of 47.1\% and 52.8\%, respectively. This research contributes to the field of occupancy detection by providing a~cost-effective solution that utilizes existing Wi-Fi infrastructure and demonstrates the potential of machine learning techniques in accurately classifying room occupancy.