Improving Location Fingerprinting through Motion Detection and Asynchronous Interval Labeling Philipp Bolliger 1 , Kurt Partridge 2 , Maurice Chu 2 , Marc Langheinrich 3 1 Institute for Pervasive Computing, ETH Zurich, Switzerland 2 Palo Alto Research Center, Palo Alto, CA, USA 3 Faculty of Informatics, University of Lugano, Switzerland Abstract. Wireless signal strength fingerprinting has become an in- creasingly popular technique for realizing indoor localization systems using existing WiFi infrastructures. However, these systems typically re- quire a time-consuming and costly training phase to build the radio map. Moreover, since radio signals change and fluctuate over time, map main- tenance requires continuous re-calibration. We introduce a new concept called “asynchronous interval labeling” that addresses these problems in the context of user-generated place labels. By using an accelerometer to detect whether a device is moving or stationary, the system can con- tinuously and unobtrusively learn from all radio measurements during a stationary period, thus greatly increasing the number of available sam- ples. Movement information also allows the system to improve the user experience by deferring labeling to a later, more suitable moment. Initial experiments with our system show considerable increases in data col- lected and improvements to inferred location likelihood, with negligible overhead reported by users. 1 Introduction WiFi localization has shown great promise for indoor positioning, yet has not achieved ubiquitous commercial success yet. One difficulty has been the con- struction of an accurate mapping between signal strength patterns and physical locations. The signal strength patterns depend not only on the distances between WiFi radios, but also on other factors such as the positions of physical objects, which reflect or block signals. This complication may be partially overcome by either performing calculations with detailed models of the environment, or by collecting a dense dataset of fingerprints and their associated true locations [2]. In this paper, we focus on the latter approach, as it is generally more accurate and it is easier to collect this data. Even so, collecting labeled fingerprint samples can be tedious. Signal readings must be collected every few meters or so, with pauses of tens of seconds at each position to get an accurate reading. This process must be repeated if the infras- tructure or environment changes substantially. Commercial deployments usually