Nine-Axis IMU-based Extended inertial odometry neural network

Abstract

With the development of mobile devices, such as smartphones, research on fast and accurate trajectory tracking is being actively conducted. This research requires a continuous integration of the acceleration and angular velocity data obtained from the low-cost microelectromechanical system-based inertial measurement unit (IMU) installed in a device to track the user’s trajectory. During this process, drift occurs over time due to the bias and intrinsic error of the IMU sensor. Hence, the 6-Axis IMU-based inertial odometry neural network (IONet) using deep learning, which is designed as a framework for velocity estimation, is used to reduce drift by dividing the acceleration data into independent windows. However, drift still occurs in estimating a pose containing both a position and an orientation because the integration of pose changes is also required. In this study, we proposed the Extended IONet that combines a 9-Axis IONet and Pose-TuningNet to improve the accuracy of trajectory tracking by compensating for the drift problem of the 6-Axis IONet. The proposed 9-Axis IONet uses the gravitational acceleration and geomagnetic data of the IMU in addition to the input structure of the existing 6-Axis IONet; thus, the estimation accuracy of pose changes improves by reducing the data dependence on the original input of the 6-Axis IONet. The proposed Pose-TuningNet is an auxiliary network that is capable of estimating pose changes more precisely using the higher-dimensional inclination-angle information obtained from the IMU to focus on the noise model of the IMU. Experiments were conducted using the Oxford Inertial Odometry Dataset, which is public dataset for deep learning based inertial navigation research to verify the performance of the proposed neural network. Compared with the existing 6-Axis IONet, the Extended IONet achieved superior performance in five out of seven cases, and its overall 39.8% RMSE improvement demonstrated its excellent performance. Additionally, the results showed that Pose-TuningNet improved the position estimation performance by correcting the drift problem in the 9-Axis IONet.