Size Does Matter - Positioning on the Wrist A Comparative Study: SmartWatch vs. SmartPhone Gerold Hoelzl and Matthias Kranz Faculty of Computer Science and Mathematics University of Passau Passau, Germany Email: forename.surname@uni-passau.de Andreas Schmid, Peter Halbmayer and Alois Ferscha Institute for Pervasive Computing Johannes Kepler University Linz, Austria Email: surname@pervasive.at Abstract—Indoor Positioning is a crucial topic to provide autonomous services to people based on their location. Nowadays dominating positioning systems, like GPS (Global Positioning System), are designed for outdoor use not applicable for indoor scenarios as they depend on a direct line of sight to reference stations. Recent progress in wearable computing peaked in the promising development of SmartWatches. They are seen as a successor of the SmartPhone evoking a new era of an always on, large scale, planet spanning, body sensor network. This work investigates in the question if SmartWatches are an accurate and suitable approach for an out of the lab, 24/7, real world Indoor Positioning System. In utilising Wi-Fi fingerprinting methodologies in combination with machine learning techniques, it is shown that state of the art consumer hardware in form of SmartWatches can be used to shape a cost effective, unobtrusive, and accurate indoor positioning system. Keywords—SmartWatches; Realtime Indoor Positioning; Wearable Computing; I. I NTRODUCTION Persons’ positions inside buildings deliver highly important contextual input information for smart services. SmartHomes, as one application scenario, can use this information to au- tonomously reduce energy consumption by turning not used electrical power consumers off [1], [2], [3]. SmartHomes can utilize the position information to e.g., autonomously switch off the lights in the kitchen when no person is in the kitchen, or to turn the lights on in the living room when a person is recognized to walk to the living room. These two simple and placative scenarios highlight the usefulness and the need for a real world, 24/7 utilizable cheap and reliable positioning system ready for daily use. The scenarios are of course not limited to electrical power consumers, and can easily be extended to other energy consumers like the com- ponents of HVAC (heating, ventilation, and air conditioning) systems. Indoor positioning systems are not limited to the use in SmartHomes. They can be used in different appli- cation scenarios like location based network access, games, logistics and security [4]. The vast majority of today’s indoor positioning systems is not utilizable for people due to their obtrusiveness, cost factors, and their complicated setup and maintenance procedures. In this work, we argue to utilize an ecosystem of Wi-Fi-Access Points in combination with SmartWatches to achieve an accurate, unobtrusive, daily usable indoor positioning system for the use in out of the lab settings. The research hypothesises valid within this paper can be formulated as follows: (i) The expected positioning accuracy of a SmartWatch is applicable for implicit SmartHome con- trol based on location information; and (ii) Compared to a SmartPhone the accuracy drain of a SmartWatch is negligible; II. RELATED WORK The basic indoor positioning techniques can be clustered into (i) dead reckoning [5], (ii) proximity sensing [6], (iii) triangulation [7], (iv) trilateration [7], and (v) fingerprinting (scene analysis) [8]. The use of Wi-Fi signals for positioning systems has become popular during the last years [4], mainly due to the increasing availability of Wi-Fi Access Points (APs). As of today, it is very unlikely in a urban area to find a spot where not at least one SSID (Service Set Identifier) is received by a Wi-Fi client. Today’s Wi-Fi based location estimation approaches use the Received Signal Strength Indicator (RSSI) from various APs to build a RSSI - Fingerprint database which is used for the positioning of the Wi-Fi clients. One of the first basic works was done by Bahl and Padmanabhan [9]. They proposed an in-building user location and tracking system named RADAR, which uses Wi-Fi signal strength data with the k-NN machine learning algorithm. The median error distance of RADAR is 2 to 3 meters, about the size of a typical office room. A slightly better accuracy can be achieved by additionally using the orientation of the Wi-Fi client, as presented by Chan et al. [10]. By using the built-in orientation sensors from an Android SmartPhone (Google Nexus One) the positioning accuracy raised to 1.82m. III. EXPERIMENT DESIGN We present that the positioning accuracy that can be achieved with a Wi-Fi based fingerprinting system fully implemented on a SmartWatch is sufficient to make positioning possible at least on room-size level in a daily usable setup. We assume that the average minimum room size is 10m 2 or at least 3 × 3m. This room size sets our accuracy goal to 3m. It was already shown that with Wi-Fi fingerprinting this accuracy is achievable. All the work which has been done on Wi-Fi fingerprinting so far shows a reliable positioning accuracy between 2 to 3m [8], [9], [10], [11]. The Third IEEE International Workshop on Sensing Systems and Applications Using Wrist Worn Smart Devices, 2017 978-1-5090-4338-5/17/$31.00 ©2017 IEEE 703