A WiFi RSSI Ranking Fingerprint Positioning System and Its Application to Indoor Activities of Daily Living Recognition

Abstract

WiFi RSSI (Received Signal Strength Indicators) seem to be the basis of the most widely used method for Indoor Positioning Systems (IPS) driven by the growth of deployed WiFi Access Points (AP), especially within urban areas. However, there are still several challenges to be tackled: its accuracy is often 2-3m, it’s prone to interference and attenuation effects, and the diversity of Radio Frequency (RF) receivers, e.g., smartphones, affects its accuracy. RSSI fingerprinting can be used to mitigate against interference and attenuation effects. In this paper, we present a novel, more accurate, RSSI ranking-based method that consists of three parts. First, an AP selection based on a Genetic Algorithm (GA) is applied to reduce the positioning computational cost and increase the positioning accuracy. Second, Kendall Tau Correlation Coefficient (KTCC) and a Convolutional Neural Network (CNN) are applied to extract the ranking features for estimating locations. Third, an Extended Kalman filter (EKF) is then used to smooth the estimated sequential locations before Multi-Dimensional Dynamic Time Warping (MD-DTW) is used to match similar trajectories or paths representing ADLs from different or the same users that vary in time and space In order to leverage and evaluate our IPS system, we also used it to recognise Activities of Daily Living (ADL) in an office like environment. It was able to achieve an average positioning accuracy of 1.42m and a 79.5% recognition accuracy for 9 location-driven activities.