Spatial Inference Network: Indoor Proximity Detection via Multiple Hypothesis Testing

Spatial inference is an important task in large-scale wireless sensor networks, the Internet of Things, radio spectrum monitoring, and smart cities. In this paper, we extend and adopt our spatial multiple hypothesis testing approach with false discovery rate control to a real-world spatial inference sensor system detecting the presence of people in indoor settings. The developed inference method is data driven, using empirical statistics and conformal p-values instead of assuming specific probability models. The approach has both, low computational complexity and energy efficient communication, hence expanding the lifespan of the network. Each sensor computes local p-values and communicates them to a fusion center. This performs the actual testing and identifies the regions where the alternative hypotheses are in place. The reliable performance of the method is demonstrated using real-world measured data acquired by an indoor wireless sensor network.