Article Preview
TopIntroduction
Recognition of human activities is frequently achieved by inertial and body-worn sensors (Yang, Xue, Fang, & Tang, 2012; Lukowicz, Pentland, & Ferscha, 2012). In some application cases, however, these sensing systems are not applicable. This touches, in particular, cases with non-cooperating persons or crowd that can not easily be equipped with sensing devices. While primarily video or audio have been heavily investigated as environmental sensing sources (Aggarwal & Ryoo, 2011; Chaquet, Carmona, & Fern´aNdez-Caballero, 2013; Kunze & Lukowicz, 2007; Schuermann & Sigg, 2013), also less intrusive RF-based systems can be employed for the localisation of entities or the recognition of activities from non-equipped individuals. In these systems, the fluctuation in the signal strength, phase or frequency of a received RF-signal is exploited. Individuals in the proximity of a receive antenna blocking or reflecting signal propagation paths might incur characteristic patterns in the signal strength propagation (Scholz, Sigg, Schmidtke, & Beigl, 2011). These systems are called device-free since the monitored entity is not equipped with any receive or transmit device (Sigg, Scholz, Shi, Ji, & Beigl, 2013; M. Youssef, Mah, & Agrawala, 2007). While the localisation of individuals has received much attention (see (Zhang, Liu, Guo, Gao, & Ni, 2012) and references therein), current work is shifting to other domains such as the counting of multiple subjects (Xu et al., 2013), the sensing of traffic situations (Ding, Banitalebi, Miyaki, & Beigl, 2011; Kassem, Kosba, & Youssef, 2012), or the distinction of activities (Sigg, Scholz, et al., 2013; Shi, Sigg, & Ji, 2012b; Patwari & Wilson, 2011). In previous studies, in particular in active installations, which incorporate a transmitter as part of the system, typically a single receiver and a single individual conducting activities at a time are investigated. We focus in this study on three relevant but yet untackled research issues in device-free recognition systems:
1 The simultaneous recognition of activities from multiple persons by single-antenna receive devices
2 The impact of multiple single-antenna receive devices on the recognition accuracy
3 The recognition of gestures from single antenna devices on time-domain features
All features are extracted from the time domain of a received signal since contemporary end-user devices are not applicable of deriving more sophisticated, such as frequency domain features. We employ USRP (http://www.ettus.com) software defined radio (SDR) devices placed in an environment for an active device-free activity recognition system. With such RF-sensing capabilities, a seamless integration of pro-active systems in smart environments becomes possible. For instance (See Figure 1), consider touch-free interaction or the recognition of activities from possibly multiple persons in a smart environment. This article is an extended version of our work published in (Sigg, Shi, & Ji, 2013). We add a comprehensive discussion of related work, further details on the simultaneous recognition of multiple activities (in particular adding precision and recall for all results) and a complete novel study on the recognition of gestures from time-domain features.
Figure 1. Utilisation of RF-fluctuation