Action Recognition Using HOG Feature in Different Resolution Video Sequences

This paper summarizes the recent development of action recognition at first. Then based on Hierarchical Filtered Motion model and Nearest Neighbor classifier, we do action recognition using HOG feature in video sequences of different resolutions. Here we use KTH dataset for training and MSR action dataset II for testing. The experiment demonstrates that the new feature extraction process is effective and has better performance in the cross-dataset action recognition.