Cross-View Person Identification by Matching Human Poses Estimated With Confidence on Each Body Joint