Experimental validation of the ResNet layer number design method for Wi-Fi location estimation in different environments

Abstract

In this study, we performed indoor location estimation using wireless LAN. The estimation method is based on the Finger Print method [1]. We measure the database (DB) and user data (UD) using wireless LAN radio waves to improve the location estimation accuracy of Finger Print indoor location estimation. The Neural Network (NN) that compares UD and DB is ResNet (Residual Network), which is a derivative of CNN (Convolutional Neural Network). The number of layers that provide the best accuracy varies depending on the environment. To confirm this, we experimentally verified the relationship between the number of layers and the estimation accuracy in different environments, and clarified the design method.