Non-Intrusive Human Motion Recognition Using Distributed Sparse Sensors and the Genetic Algorithm Based Neural Network

Abstract

Due to the rapid development of sensing technology and the increasing ratio of elderly population, many research activities have been performed to develop human motion detection and recognition systems. Various camera and wearable sensor-based human recognition systems have been developed; however, they are either not privacy protective or not practical for longterm monitoring. In this paper, we present a non-intrusive indoor human recognition system using distributed sensors and Genetic algorithm (GA) based neural network. Pyroelectric infrared (PIR) sensors are chosen using masks with random sampling windows to sense the human body thermal variations. The time domain statistical features are extracted to train classification algorithm in order to recognize human motion. Total of 200 samples are collected from volunteers performing two actions, i.e., walking normal and abnormally. A number of classification algorithms have been trained to recognize human motion. The outcome indicates that the QFAM-GA method outperforms other state-of-the-art methods, such as KNN, SVM, CART, NB and Fuzzy Min-Max.