Pyroelectric InfraRed sensors based distance estimation

In this paper a novel technique is proposed to detect person position through the use of an array of lowpower, low-cost Pyroelectric InfraRed (PIR) detectors. Typically, PIR sensing elements are used in surveillance or automatic lighting systems to provide a presence/absence digital signal. However, much more information can be extracted from sensors output. Our approach combines the output from two detectors placed on opposite walls of a hallway and facing each other. Through the fusion of simple features calculated locally on sensor nodes we are able to classify in real-time passages through the hallway into three classes according to the distance of the person from the sensors. We evaluated the use of three classifiers: Naïve Bayes, k-Nearest Neighbor (k-NN) and Support Vector Machines (SVM). We achieved a correct classification ratio of 83.49% using naïve Bayes classifier, 86.06% using a linear SVM classifier and 93.75% using 3-NN (k=3) classifier.