Abnormal Behavior Detection In Complex Scene

In recent years,the frequent outbreak of public safety incidents,it has been unable to meet the requirements of timely and efficiency for solely relying on manual inspection of public security.Therefore,it means becoming essential to improve the efficiency of anomaly detection relying on computer technology.So,the abnormal detection based on computer vision is becoming the focus of current research.Although researchers have made a lot of work in this field,however,abnormal behavior detection in complex scenes still have many challenges.Therefore,the computer vision technology,machine learning and other methods are employed for abnormal behavior detection in different scenarios videos.First,videos need to be processed before abnormal behavior being detected.According to the different quality of surveillance video,it need to be preprocessed by gray transform,morphological processing,denoising,enhancement and other image processing technology.it is pretreatment prepare for human detection in the later.Second,human targets are detected in complex scenes.The detection method of various foreground objects,including Gauss background modeling,background subtraction method,frame difference method,optical flow method and according to the human body.The characteristics of moving target under complex scene,the noise characteristics of randomness,this paper combines the advantages of mass detection and prospect of optical flow method the detection method based on visual attention model,put forward human targets in complex scenes blob detection and optical flow method the detection algorithm based on.And experiments were conducted and compared.Experimental results show that our proposed algorithm is superior to the traditional method,can effectively detect human targets.Third,visual features of human behavior are extracted based on feature fusion.According to the characteristics of complex scene video,the multi-scale optical flow histogram feature extraction algorithm is proposed based on mass and the integration of static features which are as visual features to describe human behavior.It not only maks full use of temporal information of video data and local texture information,but also has anti scaling,so as to improve the robustness to describe the characteristics of human behavior and detect abnormal behavior.Finally,the establishment of human abnormal behavior detection model under complex scene.We use multilayer recurrent neural network ability to solve nonlinear problems and establish the abnormal behavior detection model based on multilayer recurrent neural network.And the model is tested in three different scenarios of data sets,and the experimental results are compared with the results of multi-scale optical flow method based on the gray histogram.The experimental results show that the accuracies of proposed method is more better and it demonstrates that the proposed method is excelled in abnormal detection.The result shows that the proposed method can realize the detection of abnormal behavior in complex scenes.