Cross-angle behavior recognition via supervised dictionary learning

The effect of behavior recognition has been very good in a fixed angle. However they do not work well in a new angle, in order to solve the limitation of single angle, the paper adopts an effective idea to solve the cross-angle behavior recognition. We propose supervised dictionary learning for cross-angle behavior recognition, which learns a common dictionary to represent the common behavior of the same behavior under different perspectives. This makes the same behavior with similar sparse representation in different perspectives. At the same time we learn a set of characteristic dictionaries to represent the same behavior under different perspectives, so that the sparse representation of the same behavior from different perspectives is distinguished. Finally, obtain the common dictionary and the characteristic dictionary of the same behavior combined with different angles, in order that the behavior can be represented and classified. Experiments show that our proposed method can more effectively solve the cross-angle behavior recognition.