A Study of an Indoor Localization System Using Self-Rotating Inertial Sensor and Deep Learning Model

Abstract

In this paper, we propose a direction estimation method using a self-rotating inertial sensor and a distance estimation method using an LSTM-based deep learning model to improve the positioning accuracy of two-dimensional moving objects such as AGVs. By rotating the inertial sensor, the direction is accurately calculated by correcting the sensor offset. With this method, it was confirmed that the direction error was reduced from 52° to 5° in 30 minutes of evaluation in a stationary state. In addition, we developed a distance estimation method using a deep learning model based on the information from the inertial sensor. We evaluated it using an AGV on a rectangular path and confirmed that it can estimate the moving distance with an accuracy of 0.7%.