Coordinate rectification of indoor neural network localization using filters

Today, we can know our outdoor location through global positioning system (GPS). However, it is not easy to estimate our indoor location because satellite signals are difficult to reach. Therefore, we study indoor localization using wireless local area network (LAN). In this work, we attempted to rectify our position estimated using a convolutional neural network (CNN). CNN of indoor localization is based on fingerprinting, and it can only estimate pre-measured coordinates. Simultaneously, CNN is not considered relation by time series. Thus, we suggest using filters like the Kalman filter or particle filter. We can estimate the pre-measured coordinates and inter-coordinates among them using filters. Additionally, we can improve the position estimation accuracy based on the temporal dependency of the user assuming a pedestrian. Our experimental validation shows that the proposed method improves the accuracy. We evaluated and enhanced the position estimation accuracy using CNN with the particle filter. Consequently, we obtained that the mean, median, and max errors decreased 0.27, 0.24, and 4.72 m, respectively, compared with only CNN.