Pyroelectric InfraRed sensors based distance estimation

Abstract

Pyroelectric infrared (PIR) sensors are low-power, low-cost devices commonly used in ambient monitoring systems in order to provide a simple, but reliable, trigger signal in presence of people. In this work we show how we are able to estimate the position of a person using PIR detectors. Our sensor node locally extracts basic features (passage duration and PIRpsilas output amplitude) and fuses them from pairs of nodes in order to classify the passages into three classes according to person position. We tested three classifiers: naive Bayes, support vector machines (SVM) and k-nearest neighbor (k-NN). All of them can be implemented on low power, low cost devices while achieving a correct classification ratio ranging from 80% up to 93%.