| Video surveillance,as an important part of Safe City and Skynet projects,is the major means of key region monitoring and urban security prevention and control.Forgery attacks on video equipment and their content are the main security problem faced by distributed video surveillance systems.According to the accuracy and reliability requirements of video surveillance in important region,this thesis focuses on the efficient video source identification and forgery detection of video surveillance systems.The specific work is as follows: In terms of video source identification,the video fingerprint extraction method based on Photo Response Non-Uniformity(PRNU)selects GOP or only I frames.Due to the impact of video compression,the methods cannot meet the accuracy rate requirements.According to the characteristics of H.264 standard and considering the I frames and available areas of undamaged B and P frames,an improved PRNU video noise extraction and fingerprint construction method and equipment fingerprint correlation detection method are proposed to realize video source identification.The experimental results show that,compared with the method based on GOP,when the recognition rate reaches 90%,the algorithm proposed in this thesis reduces the number of frames required to extract accurate noise from 220 frames to 58 frames.The recognition rate of the source device is 95.85%.Compared with the existing video fingerprint extraction method,the recognition rate is increased by 6%.It can effectively deal with video spoofing attacks and meet the security requirements for accurate identification of video surveillance systems.In terms of video forgery detection,considering the noise correlation between adjacent video frames,the video is divided into overlapping frame groups,and each frame group is taken as a query fragment to calculate the intra-group correlation and inter-group similarity.This thesis proposes a video forgery detection method based on inter-noise correlation matrix to determine the type of video forgery and locate the location of video forgery.The experimental results show that the precision rate and recall rate of this method are both above 90%.It can effectively detect two types of inter-frame duplication forgery,and meet the security requirements of efficient detection of video surveillance system.Based on the researched technologies and methods,a video security monitoring platform is designed and implemented.The above video source identification algorithm and forgery detection algorithm are applied to the platform.It can effectively deal with the forgery of homologous and heterologous videos by using video characteristics and the inherent feature of hardware devices.The platform has been deployed and applied in the police surveillance system,which verifies the feasibility and effectiveness of the proposed method. |
PDF Full Text Request