Abstract: Faced with indoor security problem brought by illegal intrusion and the prevalent Wi-Fi technique, this paper firstly designs a novel algorithm based on abnormal sequences detection to learn users’ behavior habits to identify illegal intruder. The algorithm takes collected CSI （Channel State Information） as input signal. After denoising and segmentation, the algorithm leverages HMM （Hidden Markov Model） to learn users’ behavior habits, and then adjusts detection threshold according to the output probability value. As time passed by, the model will be more close to users’ behavior feature. Finally tailored models will be built for different users, and distinguish the intruder. Experiments have demonstrated that detection accuracy can achieve 93.4%, thus the algorithm has realized real-time correct identification purpose.

Key words: intrusion detection, Wi-Fi technique, abnormal sequences detection, hidden Markov model

摘要： 针对非法入侵带来的室内安全隐患，聚焦于目前应用广泛的Wi-Fi技术，首次设计提出了一种通过学习合法用户的行为习惯，再进行异常序列检测进而甄别非法入侵者的识别算法。对收集到Wi-Fi信号的CSI特征值进行去噪和信号分段，使用隐马尔科夫模型对用户的行为建模。根据模型输出的概率不断调整判断的阈值，使学习训练的模型随着时间的推移越来越符合用户的行为特征。实验结果表明检测准确率可以达到93.4%，达到了实时准确检测的目的。

关键词: 入侵检测, Wi-Fi技术, 异常序列检测, 隐马尔科夫模型