An Investigation Of Detection And Tracking Technique Of Moving Targets Based On Multisensor Video Fusion

With the comprehensive application of security systems and traffic monitoring systems in our life,the related requirements and problems are also rapidly increasing.A traditional security system often employs a single sensor to detect and track the targets.However,it is difficult to monitor a target from multiple angles because the single camera has a small field of view and a single view angle.Especially,an obstruction between the target and the camera will dramaticly accelerate the difficulty in the detection and tracking of the target.The use of multisensor video fusion to monitor targets at different angles not only expands the surveillance field of view and reduces dead angles of monitoring,but also extends the duration of the target monitoring.Therefore,the moving target detection and tracking technique based on the multisensor video fusion has gradually become a popular research topic in the field of security systems and video surveillance.This thesis investigates the moving target detection and tracking technique based on the multisensor video fusion,which includes three functional modules:A.Video Fusion.The Speed-up Robust Feature(SURF)algorithm is used as the theoretical basis.Then the detailed derivation about the feature extraction process of SURF algorithm is given.Considering the important effect of suture searching and image fusion method in the video fusion,three methods for the suture searching and three methods for the image fusion are discussed in detail,respectively.Meanwhile,the methods of suture searching and image fusion are compared and verified via experimental examples.In order to solve the problem of image blur caused by targets in the overlapping region during video fusion,a video fusion method that based on inter-frame difference is proposed in this paper.The inter-frame difference method that is usually used for the moving target detection is used here to accurately determine the existence of targets in the overlapping area,and decide if it is necessary to re-calculate the camera parameters,find the stitching,and redo the frame image fusion.The improved image fusion method will effectively remove the image blur.B.Video Moving Target Detection.Firstly,we introduce the most popular algorithms for the video moving target detection,including the inter-frame difference method,Gaussian of Mixture Models(GMM),and the Visual Background Extractor(ViBe)algorithm.Considering the actual requirements in this application,we choose the Vi Be algorithm as a starting algorithm for the video moving target detection.To deal with the problem of sample redundancy in the background model used by the Vi Be algorithm,and the problem of ghost image and scintillation pixels associated with the Vi Be algorithm processing for the moving target detection,an improved Vibe algorithm is proposed to quickly eliminate the ghost image and scintillation pixels.The improved Vi Be algorithm fills the sample set of background model based on the ordered weights for adjacent pixels,determined by their distances to the pixel under test.Then,the image is processed by a morphological operation to eliminate the scintillation pixels.Finally,we eliminate ghosts by foreground statistics.The effectiveness and reliability of the improved Vi Be algorithm are verified through experimental examples.C.Video Moving Target Tracking.In this effort,we reviewed three representative algorithms for the video moving target tracking,including the Tracking Learning Detection(TLD)algorithm,Compressive Tracking(CT)algorithm and Kernel Correlation Filter(KCF)algorithm.After the comparison and analysis on these algorithms,we choose the KCF algorithm as a start-up algorithm to meet the requirement of fast tracking in this effort.However,the traditional KCF algorithm is sensitive to the possible deformation of the target.An improved KCF algorithm based on the HSV color feature and HOG feature fusion is proposed to relieavate the sensitivity.The accuracy and reliability of the improved KCF algorithm for the target locating are verified through experimental examples,especially when the target is deformed.