Crowd Abnormal Behavior Detection Combining Movement and Emotion Descriptors

Abstract

At present, the detection of crowd abnormal behavior has the problem of unclear definition of anomalies and mutual occlusion. Existing researches mostly detect the appearance and behavior characteristics of individuals, extract and analyze the movement characteristics of groups. Anomaly detection is achieved through comparison of parameters and thresholds. However, the context semantics cannot be effectively utilized and the definition of anomalies cannot be consistent with the actual situation. There is a problem of disconnection between behavior characteristics and behavior description. In this paper, we use the designed convolutional neural network to extract the spatiotemporal features of the crowd, combine the motion descriptors and emotional descriptors corresponding to the features, that is to say, to describe the features in multiple directions; with the help of unsupervised deep learning models, we train normal behaviors, and use crowd psychology knowledge to conduct research on crowd situation analysis; multi-class SVM training and using it to distinguish different types of features will help to achieve the description and prediction of crowd behavior. Furthermore, through the design of modular schemes to reduce the complexity of the calculation and improve the efficiency of the algorithm. The purpose of this article is to design a new method to achieve the effective extraction of various features of the crowd and the precise identification of abnormal behaviors in complex crowds. Experiments verify that the algorithm in this paper can accurately describe the behavior of complex people.