Signal processing of pyroelectric infrared sensor for classification of multiple moving human targets

Abstract

Nowadays, with the features of low energy consumption and flexible networking, the pyroelectric sensor has been applied widely in areas such as network instruction detection or human body target tracking recognition. However, due to the deficiency of prominent biological morphological characteristics, such as fingerprint, iris or facial features, in the human target signals captured by pyroelectric sensor networks node, the precise classification and identification of multi-human targets in pyroelectric sensor networks become very difficult. In this paper, we apply the wavelet packet to multi-resolution to decompose the signals of pyroelectric infrared sensors and extract the energy variation of reconstructed signals to compose feature vectors to represent various characteristics of different moving human targets. Then, we generate a classifier using those feature vectors. Through some simulation experiments, it proves that this method not only can better extract the characteristics of human targets that are in detection zone of pyroelectric infrared sensor, but also an ideal classification and recognition result can be obtained of multiple human targets by training the support vector machine classifier with proper kernel function and parameters.