Overview: Video recognition from handcrafted method to deep learning method

Abstract

With the development of information technology, the automatic recognition of human action from video becomes a very popular research topic. In this paper, we review recent state-of-the-art of human action recognition methods in videos. First, we compare several notable handcrafted methods. Then we introduce some deep learning action recognition models. As deep learning becomes hot spot of research in recent years, more and more papers have utilized this method to explore the spatiotemporal features representation. We find that the deep learning methods outperform handcrafted methods at large scale recognition especially in cluttered background. But the networks still have much disadvantage. We expect our overview provides a fairly clear guidance for future research in this domain.