An adaptive sensor network for home intrusion detection by human activity profiling

An adaptive sensor network for home intrusion detection is proposed. The sensor network combines profile-based anomaly detection and adaptive information processing based on hidden Markov models (HMM) that allow the system to train and tune the profiles automatically. The trade-off between miss-alarms and false alarms has been studied experimentally. Several types of hypothetical intrusion have been tested and successfully detected. However, hypothetical anomalies such as supposing that a resident has fallen down due to sudden illness have been difficult to detect.