Action Recognition Based on Lightweight SlowFast Convolutional Neural Network

In this paper, we propose action recognition based on lightweight SlowFast networks. We propose a lightweight neural network architecture combined with a fast and slow network to perform temporal feature extraction and spatial feature extraction through two different channels. After fusion, the accuracy of action recognition can be increased. The experimental results show that the proposed method can achieve high-precision action recognition and reduce the computational complexity of the neural network.