Varidation of Indoor Localization Method by CNN Using RSSI

Makoto Uehara, Masashi Kimura, Hideyuki Kobayashi

Abstract: Uehara, Makoto Kimura, Masashi Kobayashi, HideyukiA Global Positioning System (GPS) is widely used as a method of outdoor localization. However, indoor localization system is required because the GPS signals are disturbed indoors. Although the fingerprinting method is a simple and inexpensive method of indoor localization, it is affected by environmental factors such as multipath effects. This paper proposes a Convolutional Neural Networks (CNN)-based localization model that consists of pooling layers for Wi-Fi fingerprinting indoor localization. These pooling layers of CNN can extract features of RSSI that contribute to the localization. Experimental results show that the proposed model reduces MAE about 16%. Furthermore, the coefficient of determination improves by 0.5 compared to the fingerprinting method.