Return to Session C5 Next Abstract

Session C5: Navigation and Positioning

A Novel Height Estimation Approach Considering Barometer Sensors for 3D Indoor Positioning
Philipp Hager, Susanna Kaiser, and Christian Gentner, German Aerospace Center (DLR)
Date/Time: Friday, Sep. 20, 8:35 a.m.

Estimating the indoor position of pedestrians or first responders has been investigated intensively by the community in the recent years. Since Global Navigation Satellite System (GNSS) signals are blocked by building construction elements, other technologies are used to determine the location in indoor or underground scenarios. Using vision-based systems can result in highly accurate positioning results. However, in many scenarios vision-based systems are not usable in dark environments. Another promising approach utilizing radio signals like Bluetooth Low Energy (BLE) or ultra-wideband (UWB) or can give meter or submeter accuracy. Radio-based systems require stationary beacons to estimate the position using trilateration or triangulation. However, most buildings do not provide already existing radio infrastructure. Nevertheless, indoor positioning systems (IPSs) play a key role in a variety of different applications, like in coordinating police forces or firefighters inside a building or observing their position. Because inertial measurement units (IMUs) can offer human 3D-odometries, its adoption is a possible answer to this issue of indoor positioning. Crucially, they don't rely on infrastructure but can be easily be worn by pedestrians. IMU-based IPSs have been intensively studied recently and show favorable results in meter or even-submeter accuracy. However, IMU-based systems often show a poor long-term stability caused by sensor bias and noise. The problem of long-term stability is addressed in a system named ‘NavShoe’ using state estimates from an Unscented Kalman Filter (UKF) as the integration algorithm for the inertial measurements. The UKF propagates the mean and covariance of the estimates, utilizing multiple sigma points around the estimated mean, and then measures the mean and covariance of the propagated points. NavShoe can be extended using a Simultaneous Localization and Mapping (SLAM) approach called ‘Footslam’ to correct drift.
In this paper we show, how exploiting the information of an additional barometer sensor leads to a more accurate 3D-position estimation and an overall more reliable positioning performance.
IMUs typically provide triaxial accelerometer, gyroscope and magnetometer data and thus give information about the movement of the local sensor frame. However, the measurements obtained from inertial sensors often deviate from the real applied motion and are biased and noisy. Using double integration to obtain position, the bias and the noise can cause an exponential drift in the position.
In order to solve this problem of drift, a zero-velocity update (ZUPT) algorithm, a zero-angular-rate update (ZARU) algorithm, and a height update algorithm are utilized. While walking, the foot alternates between a stationary stance phase and a moving swing phase. In the stance phase, the shoe has no movement and the velocity is zero.
The system detects the stance phase and applies ZUPT and ZARU as pseudo measurements into the UKF. This allows the UKF to correct the velocity error during each gait cycle, breaking the exponential-in-time error growth and also updating the accelerometer and gyroscope biases. The UKF uses a state vector consisting of 15 states with the euler angles, the gyroscope bias, the position, the velocity, and the accelerometer bias, each of them in three dimensions.
The UKF uses a measurement vector consisting of 8 entries with 3D zero velocity as a pseudo-measurement in case of a ZUPT, the estimated 3D gyro bias in case of a ZARU, the estimated accelerometer bias and the estimated height of the vertical axis in case of a ZUPT.
Considering the availability of an already in the IMU integrated barometer, the atmospheric pressure data might add valuable information to the NavShoe. Using the reference and measurement pressure, the reference height, the gravity constant, the gas constant and the thermodynamic temperature of the air, the measurement height can be determined. The height relative to the initial height can the be used as a second source of height information for UKF measurement vector.
For the evaluation of the UKF and to test the barometer performance, indoor measurements where performed. While walking, the barometer data was recorded and smoothed using a moving average filter to obtain less noise in the data. However, this measurement confirms the barometer as a valid source for height estimation, since the barometer data shows a constant height, even though the acceleration-based height changes due to falsely detected steps.
To summarize, in this paper we show a novel indoor positioning approach by extending the NavShoe system with a barometer. Contrarily to current state-of-the-art algorithms, the height estimation relies not the inertial measurements and also on barometer measurements. To the knowledge of the author, this is the first time that an UKF with two different and independent measurement sources is proposed to obtain more accurate height estimation in 3D indoor positioning. Based on indoor measurements, first favorable results where achieved where the barometer can improve the height estimation.

Return to Session C5 Next Abstract