CNN and RNN Based Neural Networks for Action Recognition
Chen Zhao, Jun Gang Han, Xuebin Xu · Journal of Physics Conference Series · 2018
In this paper, we propose a novel approach to recognize human actions. Action recognition from videos has not been addressed extensively and effectively, primarily due to the tremendous variations that result from background and scale, etc. In order to classify the videos, at first we use very deep Convolution Neural Networks(CNNs) to extract the features of the frame in videos. Then, we use multilayerd Recurrent Neural Networks(RNNs) with a type of Long Short-Term Memory(LSTM) units to process the sequence of the extracted features by CNNs. Finally, we integrate both of the CNNs and RNNs for our model. We evaluate the model on UCF-11(YouTube Action) dataset and analyze which type of the model is fit for classify the videos. And the model we trained gets a higher accuracy.