Smartphone Sensor-Based Orientation Determination for Indoor-Navigation

Abstract

Many methods of indoor navigation for smartphones are augmented with Pedestrian Dead Reckoning (PDR) to improve accuracy and to reduce latency. PDR requires an accurate estimate of the device orientation. From the pitch and roll angles the sensor readings can be rotated to the horizontal plane, and with the yaw angle the direction of movement can be determined. While a simple implementation using only accelerometer and magnetometer is possible, more accurate results may be obtained by also including the gyroscope measurements. The approach in this paper uses a Kalman filter to fuse gyroscope with accelerometer and magnetometer readings. The system equation uses random walk on straight trajectories and additional gyroscope readings on turns. Turns are detected using a statistical test on the innovation of the Kalman filter as well as a condition on the estimated yaw-rate from the gyroscope. A second Kalman filter separates gravity from specific force by processing acceleration measurements. The estimated gravity is used in the orientation filter to determine pitch and roll. The filter has been tested using trajectories with known ground truth taken with off the shelf mobile devices in corridor and office environments. The outer heading accuracy approaches 10°, dominated by systematic effects, largely due to magnetic disturbances. The achieved inner accuracy for the heading is 4°.