Inferring human activity using overcomplete dictionary based pattern recognition

Inferring human activity is often achieved using specialized sensors and beacons such as accelerometers, pedometers and motion sensors on wireless nodes. These sensor nodes compose a wireless sensor network in a coordinated way. However, these sensors are too expensive for large deployment, which affects performance. In addition, the transmission accuracy of wireless communications suffers from radio irregularities caused by obstacles such as human bodies in the environment. The human body selectively reflects, diffracts and scatters the radio signal which affects the received signal strength. This paper presents an approach to detecting human activity using fluctuations in the received signal strength. These fluctuations are detected using an overcomplete dictionary based pattern recognition algorithm. Performance results are presented which show that the proposed system has an accuracy of 86% in detecting human activity. Moreover, the detection algorithm can be implemented in software without modifying the existing infrastructure. As result, this is a promising technology for security and surveillance applications.