Dynamic Sign Language Recognition based on Depth Motion Volumes and Key Frames

Abstract

This paper proposes a dynamic sign language recognition approach through the use of
Depth Motion Volumes (DMV). Different from the previous research by stacking the
difference of the motion energy of depth images, a novel strategy is applied for
forming DMV. The selected key frames of the entire video sequences are stacked in
chronological order and then projected into three orthogonal planes. Distinctive
features are calculated by Histograms of Oriented Gradients (HOG) and then input to
LIBSVM for recognition. The experiments are executed on Microsoft Research
Gesture3D dataset and the results indicate that the proposed approach is efficient and
performs better than the compared approaches