Neural Network for Predicting Error of AP Location Estimation Method Using Crowdsourced Wi-Fi Fingerprints

Abstract

RSS values observed from a smartphone are related with distances to each AP. Therefore, AP locations can be estimated when enough number of location-labeled Wi-Fi fingerprints are obtained. Since manually collecting Wi-Fi fingerprints costs human labor, crowdsourcing approach is preferred. Crowdsourced Wi-Fi fingerprints usually need an additional step to tag a location label. The low accuracy of indirectly acquired location labels affects the result of AP location estimation. Therefore, some AP locations need to be discarded if the error of estimated AP location is high. To measure the error, it is necessary to survey the ground truth of AP location. Since surveying true AP locations also costs human labor, an error prediction method is helpful. We propose the neural network that predicts the error of an estimated AP location. The performance of the proposed method was tested on KAIST N1 building, Cheongju airport, and Lotte World mall.