An Optical-Camera Complement to a PIR Sensor Array for Intrusion Detection and Classification in an Outdoor Environment

An important issue faced while employing Pyroelectric InfraRed (PIR) sensors in an outdoor Wireless Sensor Network (WSN) deployment for intrusion detection, is that the output of the PIR sensor can, as shown in a recent paper, degenerate into a weak and unpredictable signal when the background temperature is close to that of the intruder. The current paper explores the use of an optical camera as a complementary sensing modality in an outdoor WSN deployment to reliably handle such situations. A combination of background-subtraction and the Lucas-Kanade optical-flow algorithms is used to classify between human and animal in an outdoor environment based on video data. The algorithms were developed keeping in mind the need for the camera to act when called upon, as a substitute for the PIR sensor by turning in comparable classification accuracies. All algorithms are implemented on a mote in the case of the PIR sensor array and on an Odroid single-board computer in the case of the optical camera. Three sets of experimental results are presented. The first set shows the optical-camera platform to turn in under supervised learning, high accuracy classification (in excess of 95%) comparable to that of the PIR sensor array. The second set of results correspond to an outdoor WSN deployment over a period of 7 days where similar accuracies are achieved. The final set also corresponds to a single-day outdoor WSN deployment and shows that the optical camera can act as a stand-in for the PIR sensor array when the ambient temperature conditions cause the PIR sensor to perform poorly