Research on abnormal behavior recognition algorithm and early warning in scenic spots under SlowFast architecture
Due to the lack of relevant examples,limited datasets,low level of ancient architectures,and poor quality of abnormal behavior recognition in the current scene of ancient architecture,the accuracy of abnormal behavior recognition is low.This paper independently shot multiple sets of videos in the background of ancient architectural scenic spots,and selected 5 033 video datasets of personnel abnormal behavior from them:with clear typical ancient architectural backgrounds;with characteristics of violent terrorism fights,scratch,description,and abnormal behavior of individuals with fire risks in multi-person scenario,which each video were annotated.This paper successfully introduced signal temporal feature activity and mobility parameters into the SlowFast network framework for the first time,and modeled high-order temporal features and added classification operators to the constructed dataset.In the task of abnormal behavior recognition,the Top1 accuracy of the model reached 93.54%,while the mean accuracy reached 96.30% .After introducing activity and mobility operators into SlowFast model,the accuracy of model recognition improved by 0.87% .Compared with several common architecture algorithms,the method proposed in this paper has certain advantages.
万方数据
维普
