Signal processing and pose estimation using accelerometers

Abstract

When capturing the pose of actors for use in games, commercial companies often use dedicated rooms with expensive equipment to accurately establish the pose of an actor. This may be prohibitive when the location is unsuitable or the equipment too expensive. This thesis describes accelerometers as a cheap and portable sensor and it's use to perform motion capture on humans. This was done by providing a model and functions to derive the current pose based on captured data, testing its functioning under theoretical optimal conditions, as well as a small sample of live tests involving real hardware and human subjects. The tests demonstrate that pure accelerometers have intrinsic problems, and that input restrictions or alternative measuring equipment are a better path to success.