Convolution neutral network enhanced binary sensor network for human activity recognition

Abstract

This paper presents binary sensor network for human activity recognition, specifically, whose performance is enhanced by using convolution neutral networks (CNNs). The goal of this research is to develop a sensing system based on CNNs that can recognize various human daily activities with minimum data acquisition. The whole system consists of pyroelectric infrared (PIR) sensor arrays, feature extractor, and a classifier. All the sensory data are converted into binary numbers to preserve the geometry and motion information of targets. In this work, we compare the feature selection and classification methods of CNNs, which can extract feature automatically, with another method. The experiment results have demonstrated that the proposed system can recognize human activity recognition only with a few bits of sensory data. Besides, the CNNs can achieve the best recognition performance despite their high computational complexity.