Action Recognition Using Weighted Fusion of Depth Images and Skeleton's Key Frames