Stationary and Moving Occupancy Detection Using the SLEEPIR Sensor Module and Machine Learning

Abstract

Passive infrared (PIR) sensors, currently used for indoor lighting control report frequent false negative detects for stationary occupancy, which takes up almost 50% of the total occupancy rate. To address this, we developed a synchronized low-energy electronically chopped PIR (SLEEPIR) sensor that incorporates a liquid crystal (LC) shutter to chop the long-wave infrared signal received by the on-board PIR sensor. In this paper, we present a SLEEPIR sensor module, integrated with a PIR and machine learning for systematic evaluation of true occupancy detection in daily life. We design complex experimental scenarios containing a series of continuous daily activities and individual actions to simulate realistic environment to the maximum extend. We extract and down select key statistical features using recursive feature elimination with cross validation. In the end, we compared six machine learning models to evaluate the detection performance. Experiments involving continuous daily activities indicate an accuracy of 99.12% by using the support vector machine classifier.