View-invariant Feature using Pose Information and Flexible Matching Algorithm for Action Retrieval.