View invariant hand gesture recognition using 3D trajectory

In this paper we propose a view invariant hand gesture recognition algorithm based on the assumption that the gesture trajectory is almost in a plane, which we call principal gesture plane. We use Least Squares Method to estimate the plane and project the 3D trajectory onto it. The viewpoint-dependent problem is solved by the projection. HMMs are chosen to model the gestures. We have evaluated the algorithm with two experiments from different viewpoints. The results are promising with average recognition rate at 93.4% and 93.7%.