With the improvement of safety requirements in public places, in order to realize efficient and accurate identification of dangerous behaviors, this article carries out related research based on DL (Deep learning). In this article, an optimized architecture based on CNN (Convolutional neural network) is constructed. The attention mechanism and the method combining spatio-temporal characteristics are used to extract and select features, and SVM (Support vector machine) combined with ensemble learning is used to construct a classification model. The experiment uses data sets covering a variety of scenes, and after labeling and preprocessing, it is divided into training set, verification set and test set. The results show that the accuracy of the model in identifying Falling behavior is 92%. In the whole training process, the accuracy is stable at around 88% after 50 rounds of training, the recall rate is stable at around 85% after 40 rounds of training, and the F1 value is finally stable at around 86% under different test set sizes. Complex scenes have an impact on the performance of the model. The accuracy of simple scenes is over 90%, and that of complex scenes is reduced to about 80%. The model has good performance, but it still needs to be optimized in the face of complex scenes and unbalanced data, which provides a new method and idea for identifying dangerous behaviors in public places.
Wei et al. (Sun,) studied this question.