Clustered Multi-task Linear Discriminant Analysis for View Invariant Color-Depth Action Recognition