Learning Behavior Recognition and Analysis by Using 3D Convolutional Neural Networks

Due to the diversity of human behavior, scene noise, camera view angle and other characteristics, the difficulty of recognizing human behavior has increased. Nowadays, there are more and more problems, such as low learning efficiency or excessive mental pressure. To a large extent, these problems are caused by an unreasonable distribution of work and rest. Therefore, this paper proposes a learning behavior recognition method based on the structure of 3D convolutional neural networks (CNN). This involves taking several continuous frames of video as a group, through effective training, after the convolution and pooling operation to extract the action information in the features. Finally, the recognition and classification results are obtained through the full connection layer and the classifier. The experimental results show that this method is accurate and fast. Taking the KTH dataset as an example, the methods of improving test accuracy under different parameters were compared and discussed, which provided a theoretical basis for learning behavior dataset recognition.