ORIGINAL ARTICLE Redundancy reduction for indoor device-free localization Jinjun Liu 1,2 • Ning An 1 • Md. Tanbir Hassan 1 • Min Peng 1 • Zheng Cui 3 • Shenghui Zhao 2 Received: 1 September 2016 / Accepted: 30 September 2016 Ó Springer-Verlag London 2016 Abstract To improve localization accuracy, device-free passive localization studies usually deploy a number of sensor nodes in indoor environments, which causes redundant features and produces large data volumes and high deployment costs. This paper proposes the concept of a two-level redundancy and formulates the node reduction problem as a redundancy control problem. With the goal of using fewer nodes while maintaining high localization accuracy, a method is proposed to control the two-level redundancy efficiently and reduce the number of nodes greatly. Experiments are performed in two completely different environments. The proposed method is able to maintain accuracy levels above 90% and can efficiently reduce the total number of nodes by 59.09% in a large room (150 m 2 ) and by 68.75% in a small room (25 m 2 ). Furthermore, due to reduced nodes the proposed method can drastically reduce the needed amount of localization data and the hardware costs. Keywords Indoor passive localization  Redundancy reduction  Node optimization  Node reduction  Reducing the amount of data 1 Introduction With the development of ambient intelligence (AmI), an increasing number of wireless sensors have been intro- duced to indoor environments, including homes, offices and public places. These sensors generate sensor-based big data and bring new demands and applications [1–5]. One of these applications is the indoor device-free passive (DfP) localization technique that uses radio frequency (RF) sig- nals from these wireless sensors. DfP was first proposed by Youssef [6], and it senses an entity that is not carrying any devices by analyzing the disturbance caused by the human body to the received signal strength (RSS) values of mul- tiple WiFi links. Over the last decade, many studies have been conducted on DfP techniques, including fingerprint localization [6–8], radio tomographic imaging [9, 10], multi-person localization [11, 12], and finer-grained chan- nel state information (CSI)-based localization [13]. To improve the localization accuracy, DfP methods usually deploy too many nodes in indoor environments. These excessive nodes increase the hardware costs, local- ization data volumes, localization time, packet collisions, etc. More importantly, from the view of the feature selec- tion, these excessive nodes result in the two-level redun- dancy, thereby conversely affecting the localization accuracy. The first level is the node redundancy caused by the excessive nodes, and it causes excessive links (a wireless link is composed of a transmitter and a receiver and is also a feature in the localization dataset). The second level is subtle and occurs when a person appears in an indoor position point and disturbs the multiple links around the point; this disturbance causes uniform changes to RSS values from these links and leads to a high correlation among the disturbed links (i.e., link redundancy), even in the case of no excessive nodes. In short, the two-level & Ning An ning.g.an@acm.org 1 School of Computer and Information, Hefei University of Technology, Hefei, China 2 School of Computer and Information Engineering, Chuzhou University, Chuzhou, Anhui, China 3 Everjoy Senior Home, Hefei, Anhui, China 123 Pers Ubiquit Comput DOI 10.1007/s00779-016-0979-8