Online Learning Techniques for Occupancy Detection on Resource Constrained Devices

Our research aimed at boosting energy efficiency in HVAC systems by integrating smart technologies for dynamic control based on occupancy detection, prioritizing privacy by avoiding visual and audio data capture. We addressed cost by utilizing existing sensors and hardware, proposing a cost-effective solution for improving HVAC systems. We introduced a novel framework to enhance online learning in the IoT domain, proving its effectiveness in resource-limited scenarios. Our approach outperformed traditional batch learning, especially in dynamic settings, by maintaining high predictive accuracy even with environmental changes. Designed for resource-constrained systems, our model fits well with current IoT infrastructures. The successful simulated and real-world tests highlight the potential of our approach for broader IoT applications. This research pioneers future advancements in HVAC control by developing a reliable method to detect room occupancy, forming the basis for more advanced systems. One of the most interesting future developments is the possibility of creating an additional layer on top of this work, that can be used to predict the future occupancy of the room, allowing the HVAC system to predictively control the temperature and the ventilation, reducing the energy consumption and increasing the comfort of the occupants.