Relative Heading Estimation for Pedestrians Based on the Gravity Vector

Abstract

Inertial navigation of pedestrians carrying a smart device is a core component of many indoor positioning systems. While infrastructure-based solutions typically depend on an installation of dedicated hardware, inertial navigation depends only on sensors embedded in the device itself. A single solution can thus be applied to a large range of use cases. This work focuses on one of the main challenges in inertial navigation: user heading estimation. We describe a complete statistical model for heading estimation based on the IMU and magnetometer, assuming a fixed device pose on the pedestrian. Our aim is to provide a stand-alone solution, suitable for direct implementation into a larger positioning framework. The method consists of two consecutive parts. The first focuses on gravity vector estimation based on IMU data. We describe a method for obtaining independent estimates under dynamic conditions, thereby removing the quasi-static initialization phase required by conventional methods. The second part combines the gravity vector with gyro and magnetic measurements to estimate user heading. The proposed method is tested against a motion capture system, and against an alternative method based on attitude. We find that both methods produce similar results in terms of accuracy.