Kernel-based representation for 2D/3D motion trajectory retrieval and classification

Survey on 3D Hand Gesture Recognition

“…The hand gesture was represented by the segmented series of movements in the form of Motion History Images (MHI). Besides, more discriminative features were also proposed by projecting the motion trajectory to the higher-dimensional feature space [143]. Beh et al [144] further composed the hand motion trajectory as a unique series of straight and curved segments.…”

“…The maximum correlation coefficient was also used for the gesture recognition on Chalearn dataset [134]. For complex gesture features [143], the straightforward nearest neighbor (1-NN) with the Euclidean distance has also shown the powerful discriminative potential.…”

Combining features for Chinese sign language recognition with Kinect

“…Lin et al [17] proposed a novel kernel-space representation for motion trajectories. Kernel Principal Component Analysis (KPCA) projects a trajectory to an implicit mapping to a much higher-dimensional feature space.…”

“…Preprocessing 1) Resample the trajectories points: As different trajectories have different points number, we should resample all the trajectories to the same given L sampling points along a 3D trajectory. We adopt the method mentioned in [17]. Similarly, the length of a normalized trajectory is set to twice the averaged length, in this paper, our points number is 40.…”

“…Considering that the sign language in this work is mainly the movement of the right hand, we choose the hands' skeleton joint as the location feature. To make sure the location and scale invariant for the hand movement, we normalize the trajectories points value to a common domain as [17]. The normalized coordinates are given by…”