UC San Diego

Current methods of full body tracking have a steep correlation between hardware infrastructure and accuracy of the estimated full body pose. While implementations exist to allow for high accuracy body poses, they often require high setup overhead involving high accuracy sensors placed across all body ligaments or marker based solutions. To this end, there are existing implementations to approximate the accuracy of conventional full body tracking by utilizing sparse sensors, allowing for lower setup overhead. In this thesis, we evaluate the adeptness of Physical Inertial Poser (PIP) in accurately reproducing body poses through a mixture of single ligament and typical movements used for full body tracking. We determine that current body pose estimation solutions have a tendency to provide inaccurate results when compared against corresponding Vive tracker data. These inaccuracies are a result of faults in using accelerometers for body pose estimation, which involves sensor accuracy, limited information to solve the body pose system, and accumulation of error over time. We also propose that current body pose estimation methods could be made more accurate by measuring intermediate distances between each tracker, specifically with ultra wide-band distance tracking. This would allow the body pose estimation to take into account the correlation between each tracker and body poses, especially with regards to the root origin of the body pose estimate.