Human Behavior Recognition in Complex Scenes Based on Deep Learning

In order to overcome some of the problems that have been encountered in the past, for example, due to their reliance on manual feature extraction and limitation of model generalization, this paper proposes a new method to identify people’s behavior in complicated situations. Based on Convolutional Neural Networks (CNN), this method has been proposed for the automatic extraction of a large number of datasets. In addition, the Long Short Term Memory (LSTM) network is used to capture the long-run dependence in time order. Lastly, we use the soft max classifier to classify the various actions of people. Experiments show that the CLT network is able to achieve a high performance of 97.5% over 13 different types of people, outperforming CNN alone, LSTM, and BP models on the DaLiAc dataset, demonstrating superior performance in human behavior recognition and classification. The accuracy, recall, and F1 score evaluation indicators of the CLT net model are the highest, while all indicators of the BP model are the lowest, indicating that the CLT net model has good stability and reliability in recognizing and classifying different human behaviors.