Project Software

Body Localized Motion Capture: MocapLocalizer.java One improvement made on the motion capture data is the addition of Locap, which uses only the top six motion capture points on the dog to determine body position and orientation. Currently the motion capture determines body pose using all of the mocap points, making the position susceptible to the flakier leg mocap points. The top six points are seen by more cameras and therefore generally more reliable, yielding a better measure of body pose. In addition, the algorithm is robust to lost body points and flipped body points (when the mocap system momentarily switches the order of the points).

From the motion capture definition of the little dog robot, we can get a template of the 6 mocap markers in relation to the body center point at 0,0,0 position and orientation. Given any three of these points in the world (as observed during a run), we can find a transformation from the template trio (A'B'C') to the world trio (ABC). First define the centroids P and P' and the unit vectors shown below.

The rotation matrix from template space to world space is then

translating the space to make the centroids coincident yields the transformation matrix

This 4x4 matrix can then be applied to the original template body center to get the position and orientation of the body according to the measured world trio. This algorithm is repeated for every set of three in the 6 body points, and the largest clump is averaged to yield the final Locap pose. The number of points in the clump can then be used as a measurement validitiy check.