PIR-sensor based human motion event classification

In this paper, we use a modified passive infrared radiation or pyroelectric infrared (PIR) sensor to classify 5 different human motion events with one additional “no action” event. Event detection enables new applications in environments hosting dynamic processes. Typical event detection applications are based on audio or video sensor data. Given a data stream, often the task is to find or classify specific dynamic processes. Most of the applications for the monitoring of human activities in an environment are based on video sensor data. As an alternative or complementary approach, low cost PIR sensors can be used for such applications. The classification is done by a Bayesian approach using conditional Gaussian mixture models (CGMM) trained for each class. We show in experiments that using PIR-sensors, different human motion events in a room can be successfully detected.