Hidden Markov Model for Improved Ultrasound-Based Presence Detection

Adaptive lighting systems typically use a presence detector to save energy by switching off lights in unoccupied rooms. However, it is highly annoying when lights are erroneously turned off while a user is present (false negative, FN). This paper focuses on the estimation of presence, using a Hidden Markov Model (HMM) in a ultrasound-based presence detection system. Our results show that estimating the Log Likelihood Ratio (LLR) of presence / no-presence in real-time can achieve improvements in the accuracy of presence detection. We compare the performance of the LLR algorithm with previous presence detection algorithms. Moreover we use the concepts of receiver operating curves and a genius (perfect) detector to benchmark the trade-off between energy consumption and user comfort.