Design of a Low-Cost and Device-Free Human Activity Recognition Model for Smart LED Lighting Control

Human activity recognition (HAR) constitutes an integral part of occupant-centric smart services, such as health monitoring and building energy management. In this article, a simple and cost-effective solution to HAR through passive sensing of WiFi channel state information (CSI), is proposed. WiFi CSI extracted from ESP32 was utilized to classify four different human activities, using ensemble machine learning models. A mean accuracy of 83.39% was achieved using gradient boosting classifier with Haar wavelet-based denoising, in spite of using a single transmission link and in the presence of coexisting wireless devices. The proposed model can be employed for the development of an IoT-enabled smart LED lighting system, with minimal infrastructure changes. In this strategy, the illuminance level of LED lighting fixtures is adjusted according to the predicted occupant activity, thereby reducing power consumption, without compromising visual comfort. This method offers a potential energy savings of up to 36.42% and 29.45% per month, for a typical office and home scenario, respectively. The possibility of supplementing activity sensing with daylight harvesting, is also explored. An additional energy savings of up to 18.26% may be obtained using this method, during daytime. An evaluation of the annual electricity cost shows an estimated reduction between 29.43% and 62.13% using activity recognition and daylight harvesting.