Diversified shared latent structure based localization for blind persons

Abstract

Indoor localization systems for blind person aims to help visually impaired people localize themselves in indoor environments. Most approaches employ the RGBD camera and LIDAR for accurate localization, yet these devices are not cheap and portable for blind persons. Instead, WiFi signals are quite ubiquitous in most indoor areas, like shopping mall, hospital etc. Therefore, we propose a diversified shared latent variable model that exploits the availability of WiFi for localization. More specifically, the observation spaces in our model, WiFi strength measurements and their corresponding locations, share a single and reduced dimensionality latent space. By building and incorporating a kernel based diversity prior, the learned latent variables are inclined to extract more features of the WiFi signals, such as the coverage area, and thus further enhance the accuracy of localization. The experimental results illustrate our proposed model is accurate and efficient for indoor localization issue.