Fingerprint localization scheme with correction for missing values in training data and data augmentation

Abstract

This study discusses an indoor localization method using the radio signal strength indicator (RSSI) of a wireless local area network (LAN). An indoor localization method that adapts a convolutional neural network (CNN) to the fingerprint method is used. In this method, the CNN learns the access point (AP) information for each coordinate using the RSSI and media access control (MAC) addresses obtained from the wireless LAN APs and compares them with the AP information received from the user to estimate the user location. However, data collection for learning is costly when using CNNs. In addition, there is a problem of missing data owing to various factors when collecting AP information. Therefore, data augmentation is proposed as a method to reduce the cost of data collection while maintaining accuracy and is performed after correcting for missing values. However, data augmentation can produce unrealistic data. This paper proposes a method for correcting missing values in measurement data as a solution to this problem.