This paper presents a method for computing position and attitude of an instrument attached to the human body such as a handheld or head-mounted video camera. The system uses two Inertial Measurement Units (IMUs). One IMU is part of our earlier-developed Personal Dead-Reckoning (PDR) system, which tracks the position and heading of a walking person relative to a known starting position. The other IMU is rigidly attached to the handheld or head-mounted instrument. Our existing PDR system is substantially more accurate than conventional IMU-based systems because the IMU is mounted on the foot of the user where error correction techniques can be applied that are unavailable for IMUs mounted anywhere else on the body. However, if the walker is waving a handheld or head-mounted instrument, the position and attitude of the instrument is not known. Equipping the instrument with an additional IMU is by itself an unsatisfactory solution because that IMU is subject to accelerometer and gyro drift, which, unlike in the case of the foot-mounted IMU, cannot be corrected and cause unbounded position and heading errors. Our approach uses transfer alignment techniques and takes advantage of the fact that the handheld IMU moves with the walker. This constraint is used to bound and correct errors by a Kalman filter. The paper explains our method and presents extensive experimental results. The results show up to a five-fold reduction in heading errors for the handheld IMU.
展开▼
