Search

This work addresses the use of 3D point data to measure rigid motions, in the presence of occlusion and without reference to a prior model of relative point locations. This is a problem where cluster-based measurement techniques are used (e.g. for measuring limb movements) and no static calibration trial is available. The same problem arises when performing the task known as roving capture, in which a mobile 3D movement analysis system is moved through a volume with static markers in unknown locations and the ego-motion of the system is required in order to understand biomechanical activity in the environment. To provide a solution for both of these applications, the new concept of a visibility graph is introduced, and is combined with a generalised procrustes method adapted from ones used by the biological shape statistics and computer graphics communities. Recent results on shape space manifolds are applied to show sufficient conditions for convergence to unique solution. Algorithm source code is available and referenced here. Processing speed and rate of convergence are demonstrated using simulated data. Positional and angular accuracy are shown to be equivalent to approaches which require full calibration, to within a small fraction of input resolution. Typical processing times for sub-micron convergence are found to be fractions of a second, so the method is suitable for workflows where there may be time pressure such as in sports science and clinical analysis.

We present a novel method for dynamic estimation of pose of multiple people using multiple video cameras. Tracking is performed using a model-based approach and a set of cues which exploit both shape and colour information. For shape we propose a fast line-search method to incorporate multi-view constraints without the computational overhead of a voxel representation. The tracking algorithm is a new hierarchical stochastic sampling scheme. Results are presented using natural movements of up to two people sharing the same capture volume. Tracking is shown to be robust over a range of natural movements, including considerable occlusion. Processing times for tracking two people are at least as short as those for tracking one person using other stochastic schemes. The high performance and efficiency are attributed to the hierarchical search method and the accuracy of the cues in identifying suitable poses.