Three-dimensional gesture comparison using curvature analysis of position and orientation

This paper describes a new analysis method dedicated to the comparison of human gestures. The orientations and the positions of the gestures are first digitized using active 3D sensors and then compared to a 6-D template using curvature analysis. The proposed algorithm first starts by computing the invariant curvature of 3D position and orientation of a surgical tool using Frenet-Serret frames in 3D and quaternion space. The resulting curvature calculation is matched and compared to the template using a Dynamic Time Warping method. The proposed method is invariant to sensor position and orientation. An experimental study shows the efficiency of the new algorithm for an application in obstetrics, where the aim is to compare forceps blade placements between a senior medical doctor and a novice.