The effect of multimodal data sources in parallel on the accuracy and reliability of optical motion capture of human subjects

摘要

Optical motion capture is a powerful tool for human motion analysis. However, an intrinsic limitation of this method is its dependence upon line of sight, which when occluded, prevents measurement of position, resulting in data loss and failure of trajectory tracking.ududThe thesis aims to investigate whether sources of inertial measurement can be used in tandem with passive optical motion capture data, to obtain improvement beyond conventional measures in tracking and occlusion recovery of human motion data.ududA novel method is developed, where a conventional optical motion capture system is augmented with miniature inertial sensors. These two measurement modalities are integrated via a forward kinematic and sensor model, which facilitates the reconstruction and tracking of occluded marker trajectories.ud udThe method is validated using “gold standard” trajectory data obtained experimentally from a human subject. By synthetically degrading this data using a model of line of sight, the reconstructor is tested over a range of conditions conducive to occlusion. ududFollowing testing of trajectory data with up to 30% loss to occlusion, the method is shown to be capable of tracking marker position with 100% reliability. Additionally, trajectory data is recovered with useful accuracy (average ~8mm), demonstrating an improvement by a factor of ~10 over conventional interpolation methods. A restriction of the method is its reliance upon future trajectory data for reconstruction of occlusion, which prohibits real-time applications. udud