A Summary of Abnormal Behavior Detection for Populations

As a key research content in the field of computer vision and intelligent monitoring, population abnormal behavior detection plays an important role in maintaining public safety by analyzing video or image data. This paper systematically explains the technological development of population abnormal behavior detection, introduces traditional machine learning methods and feature extraction methods based on deep learning, focuses on analyzing the population abnormal behavior detection methods based on 2D convolutional neural network (CNN), 3D CNN and dual-stream convolutional network models, and summarizes common data sets such as UCF-Crime and ShanghaiTech Campus, as well as evaluation indicators such as AUC and accuracy, and conducts a comparative analysis of the performance of various algorithms. In response to the current challenges facing the abnormal behavior detection technology of populations, such as high computational complexity and insufficient long-term modeling capabilities, the possibility of cross-scene adaptive models, lightweight space-time modeling, multimodal fusion and other models becoming future research directions is discussed, aiming to provide more comprehensive technical references for relevant researchers in this field and help further development in this field.