Nonnegative Tensor-Based Linear Dynamical Systems for Recognizing Human Action from 3D Skeletons

Algorithm 1

Learning the nLDS model with a second-order tensor time series.

Input: Third-order non-negative tensor

Output: Parameters C and A in the nLDS model

1: Unfolding along three modes

2: Building the nLDS model of

3: Rewriting using nLDS

4: Using NTD to factorize

5: the 3-mode matricization of

6:

7: and

8: Finding the best estimate of the parameter