The Establishment And Research Of Abnormal Behavior Model In Video Image

In recent years,with the continuous development of computer vision technology in intelligent monitoring,the detection and analysis of abnormal behavior has become a hot issue in today's research.Although some research results have been obtained,it is not fully mature.This technology is still a difficult problem in the field of computer vision research.Therefore,this topic is devoted to in-depth studies of these common and widely applied abnormal behaviors such as running,tailing and fighting.First,in order to ensure the accuracy of target extraction,a non-parametric random sample model with gradient correlation is proposed to ensure the accuracy of target extraction.This method can improve the initialization of the background,update the speed of the background model,and suppress the "ghost shadow".Considering that key frame technology can reduce the workload of video analysis,an adaptive unsupervised clustering fusion cross entropy extraction algorithm is selected.With the change of the scene,the corresponding number of key frames are selected automatically,thus the limitation of the fixed threshold to the clustering algorithm is solved,and the problem of the key frame extraction is inaccurate.Second,a key frame fusion algorithm based on amplitude direction histogram entropy is proposed.The algorithm uses the adaptive unsupervised clustering fusion cross entropy key frame extraction algorithm to obtain the key frame image of the motion video,collect the histogram of the amplitude of the moving target,add entropy,and use the entropy value as the general feature to judge the abnormal behavior,thus reducing the computational complexity,and combining the key frame with the support vector machine for abnormal behavior detection.The algorithm of abnormal behavior detection is compared and analyzed.Third,in different scenes,we analyze the key motion parameters of collecting abnormal behavior,establish the abnormal model of running,trailing,and fighting,and give the flow chart of the model judgment,and measure the performance of the model from the two aspects of the recognition rate and the consumption time.The results show that the anomaly detection model can detect specific abnormal behaviors effectively.To sum up,the non-parametric random sample model which selects gradient information in the detection of human abnormal behavior can solve the problem of imprecise target detection.The adaptive clustering fusion cross entropy key frame extraction algorithm can choose the key frame adaptively,and the abnormal behavior detection algorithm is combined with the entropy of the amplitude direction histogram entropy.For key frame and SVM and Markov models,the recognition rate of abnormal behavior detection algorithm is higher,and the established model can accurately identify specific behaviors.At the same time,the efficiency of intelligent monitoring is improved,which plays a crucial role in the stability of public safety.