RSSI-Based Passive Localization in the Wild, At Streetscape Scales

Abstract

Pedestrian mobility data is valuable to data-driven decision-making for city planning, emergency response, and more. Thanks to the ubiquity of Wi–Fi-enabled devices, pedestrians may be colocalized with their devices using Received Signal Strength Indicator (RSSI) measurements from Wi–Fi probe requests, passively and privately. While shown to be feasible in controlled outdoor environments, few have used this method outdoors in production environments. In this article, we continue the work on the Mobility Intelligence System (MobIntel) and apply RSSI-based passive localization on data collected from the 500 and 400 blocks of Clematis Street in West Palm Beach, FL. We present an open-source dataset used in our study, which, to the best of our knowledge, is the first public Wi–Fi RSSI dataset for localization purposes in an outdoor environment. We then introduce a three-stage localization model that first classifies a test sample to a city block, followed by a sidewalk within the city block, and ends with an estimation of x-coordinate within the sidewalk. While we formulate the problem and validate our solution within an outdoor context, the work is equally applicable to large indoor environments. It achieves a mean localization error of 3.16 and 4.21 m, with 73% and 66% chance of reaching an error $\le$4 m, and 17% and 21% of the data discarded due to poor quality in the 500 and 400 block, respectively. We also highlight the challenges when dealing with real-world RSSI data, analyze the model's tolerance to missing data, and propose solutions to improve localization performance.