Marker-less human motion analysis is currently a hot topic in the research community. In this study three dimensional motion of a human limb is estimated using a large number of matched skin texture image patches. These two dimensional matches are triangulated and also matched to the next time frame. With these matched three dimensional points a Least Squares estimate of the rigid body motion obtained using standard methods. This motion estimate is subsequently compared to a marker based estimate acquired from a synchronized marker system.
The results show that this approach can be used for motion estimation but with less accurate results than the marker based system that is to be considered as the clinically used standard. However, the correlation surfaces indicate that the method has potential if for example subpixel correlation algorithms were to be employed.
Available as PDF (2.58 MB, no cover)
Download BibTeX entry.