Research On Fine-grained Abnormal Behavior Detection Algorithm Based On Generative Adversarial Networks

In recent years,with the increase in population density and the frequent occurrence of crowds,many casualties have occurred worldwide,seriously endangering the personal safety in public places.Accidents caused by crowding and trampling are common.For example,bustling commercial streets,stadiums with tens of thousands of spectators,and stations with frequent personnel movement are high-incidence places for crowds,and the safety issues in public places are becoming increasingly prominent.Therefore,the research on abnormal behavior detection has far-reaching practical significance.It can not only ensure the safety of crowd activities,but also provide important guidance information for crowd management and scheduling in public places.This paper learns the existing research methods of abnormal behavior detection,and puts forward an improvement plan and in-depth research on the research methods based on generative adversarial networks.The main research contents of the paper include:First of all,in order to ensure the training results of the abnormal behavior detection network,we need a data set with positive and negative samples as balanced as possible.After learning the existing pedestrian data set,I found that most of them have data imbalances.The proportion of abnormal sample is very small,and the large difference between positive and negative samples often leads to a poor learning effect of the network on abnormal behavior,which affects the accuracy of detection.Therefore,this paper proposes a method to enhance imbalanced data based on Cycle GAN.It can generate abnormal behavior data into different scenarios based on generate adversarial network,so that the proportion of positive and negative samples tends to be balanced.At the same time,the network has increased abnormal behavior consistency loss function to enhance the effectiveness of abnormal behavior data samples.In addition,current abnormal behavior detection research has different definitions of abnormal behaviors,and the network parameters are complex.It is difficult to achieve better results by adjusting parameter values.Therefore,in order to improve the accuracy of abnormal behavior detection,this paper proposes a fine-grained abnormal detection method,which not only divides the detection task into four sub-modules,but also incorporates a behavior recognition model into the detection network to refine the classification of abnormal behavior.Therefore,the fine-grained vision is realized,and finally the goal of improving the detection accuracy is achieved.This paper analyzes and improves both the data and the algorithm,and verifies the improved model on the standard dataset.The experimental results show that the improved model proposed in this paper is practical and effective,improves the accuracy of abnormal behavior detection.