Multiscale Detection of Gesture Patterns in Continuous Motion Trajectories

摘要

We describe a numerical method for scale invariant detection of gesture patterns in continuous 2D motions. The algorithm is fast due to our rejection-based reasoning achieved using a new set of curvature-based functions which we call Integral Absolute Curvatures. Detection rates above 96% are reported on a large data set consisting of 72, 000 samples with demonstrated low execution time. The technique can be used to automatically detect gesture patterns in unconstrained motions in order to enable click-free interactions.