Exploring 2D Representation and Transfer Learning Techniques for People Identification in Indoor Localization

Abstract

Indoor localization is a crucial aspect of various disciplines in our daily lives. It enables efficient administration tasks and improves safety by identifying the position of items or people inside spaces, making it useful for activities like interior navigation, asset tracking, people rescue, and building security. However, traditional systems have limited performance due to various phenomena. In this paper, a novel system is proposed to identify users inside a building using a transfer learning algorithm and a received signal strength indicator signal as an image. The system utilizes pre-trained models and the scalogram technique to increase the performance of localizing the converted data RSSI to an image. The results demonstrate that the two models can recognize users with 90% accuracy for GoogleNet and 86% accuracy for SqueezNet model.