| In modern society,public security problems of various large crowds have been paid more and more attention.The stampede in Shanghai,the Bund,sounded the alarm for crowd safety monitoring again in December 31,2014.Although the intelligent video surveillance technology in recent years has been widely used,but the crowd scenes due to the influence of dense and complicated movement and other factors,the traditional method based on the analysis of the target is difficult to effectively use.Therefore,we need to study efficient video analysis and anomaly detection methods for crowd surveillance scene,in order to effectively monitor the dynamic scene of the crowd,and avoid all kinds of security incidents.According to the traditional method of data analysis of population characteristics of high dimension,complicated model,the negative samples demand higher problem in sparse representation based on redundant dictionary framework,from the three aspects of detection,pedestrian monitoring scene crowd motion analysis and chaos of unusual events detection to carry out related research work,including the follows aspects:A pedestrian detection algorithm based on HOG redundant dictionary is studied and implemented.Monitoring the presence of pedestrians in the scene has important implications for subsequent processing.Many pedestrian free scenes are often overlooked in various analyses.This paper carried out the research of pedestrian detection algorithm,using the Histogram of Oriented Gradientsmotion feature information of monitoring scene described by sparse representation based on redundant dictionary of HOG features are modeled and analyzed to detect pedestrians within the scene,so as to determine the types of scenes,taking corresponding preventive measures can lay the foundation for the existence of pedestrian scene monitoring in the practical application.The experimental results show that this method can monitor the scene of pedestrian detection,and its performance is better than traditional pedestrian detection algorithm based on HOG feature and support vector machine,the detection rate increased by 3%,and at the same time,the detection error rate is reduced by 2%.An algorithm based on Histogram of Optical Flow is studied to analyze the crowd movement state,people with chaotic movements often have greater security risks.Therefore,effective detection of the movement of the crowd has great significance to prevent all kinds of security incidents.Aiming at this problem,using HOF feature modeling motion in the scene,the degree of confusion is described by the motion vector direction of entropy,and introduces the motion state of chaotic motion state index ? discrimination within the scene,to achieve the effective analysis of population movement.A crowd anomaly detection algorithm based on weighted multi-scale optical flow histogram and sparse representation is designed and implemented.Based on the motion direction and intensity information of the scene,a weighted multi-scale optical flow histogram model is constructed to describe the crowd movement.Then,the abnormal events in the scene can be detected effectively based on the sparse representation framework of redundant dictionary.By training redundant dictionary based on routine events,the WMHOF feature of any two frames in the scene can be sparse in the redundant dictionary.Because the abnormal event does not belong to the redundant dictionary,it is difficult to obtain sparse results.Accordingly,by defining the corresponding discriminant function,the abnormal events can be detected effectively.The experimental results show that the performance of this paper's method is better than the traditional method based on optical flow and method based on multi-scale optical flow histogram the detection of various scenes,motion intensity weighted play an effective role in detection.At the same time,because the method does not need negative samples to participate in the training process,it has more practical significance. |
PDF Full Text Request