Intrusion detection mechanism and experimental study based on WiFi signal

Abstract

Indoor safety concerns the safety of people′s life and property. Through indoor intrusion detection, it is possible to achieve early warning and avoid loss. Different from the common intrusion detection methods, the CSI(Channel status information) of the wireless communication signal WiFi is used to correlate with human behavior, which can achieve intrusion detection. The relationship between channel impulse response, channel frequency response and CSI is studied. The association between CSI and human behavior is verified using CSI dataset EHUCOUNT and machine learning method. The results show that the accuracy of intrusion detection based on SVM(Support Vector Machine) and CNN(Convolutional Neural Network) in six typical scenarios is 93.35%～99.23% and 89.17%～99.14%, respectively. Actual tests are carried out through collecting the WiFi signal using self-made spectrum sensor nodes. Experiment results show that intrusion detection accuracy is about 98%, which indicates that the intrusion detection based on WiFi signal is applicable to real scenario.

Keywords

• intrusion detection
• singular value decomposition
• machine learning
• channel state information