Mobile targets region-of-interest via distributed pyroelectric sensor network: Towards a robust, real-time context reasoning

We have established a multi-walker recognition/tracking testbed based on low-cost pyroelectrc sensor network (PSN). In order to identify a region of interest (Rol) in the monitoring area for the detection of any interesting mobile targets, we propose to use Bayesian machine learning and binary signal projection to extract the statistical contextual features from real-time, high-dimensional PSN data. This paper describes our recent results in this area, which include two aspects: (1) we have proposed to use binary principle component analysis (B-PCA) to interpret the relationship between observed sensor data and hidden context patterns. (2) We have conducted comprehensive experiments from real PSN sensor data to verify the context detection accuracy based on B-PCA models. Our results show that B-PCA can better capture context basis than general PCA algorithm.