Research On Image Mosaicing And Target Tracking For Video Surveillance

With technology development, more and more people tend to use videosurveillance instead of doing surveillance by person. Both key areas in publicplaces or private secret places are full of various video surveillance systems.However, most surveillance cameras are fixed, so the pictures obtainedrelatively have limited field of vision. If only one camera is used, a blind spotarea will appear, leaving a security risk. Development of image stitchingalgorithm is a boon for video surveillance, we can mosaic multiple cameras toa monitoring system which has a wide-angle field of view.In this paper, a research on image mosaicing algorithm assessmentmethods and target detection and tracking for video surveillance is done,which also design and implement an integrated software system. The specificcontents are:1. The ORB algorithm is selected as an image feature extraction andmatching algorithms by its real-time and accuracy. We propose two sets ofprogressive improvement program: one is based on adding scale invariance and uniform grid feature point strategy, one add the affine invariance for lowoverlap areas and high inclination application, named AORB algorithm.Performance comparison experiment shows, improved ORB has betterperformance compared to other algorithms. What’s more, the improved ORBimage mosaic algorithm not only show superior performance in the imagemosaic tests, but also does well on video splicing work.2. A complete evaluation indicators system based on analytic hierarchyprocess is constructed in this paper. The formula of registration error (RE) isimproved from TRE. Meanwhile a blind evaluation approach related on theindicators system is proposed to avoid any manual interventions and keep thecomputational simplicity. Exhaustive testing based on two algorithms has notonly shown that the proposed approach can reflect the performance of imagemosaic algorithm comprehensively and be consistent with subjectiveevaluation in results, but also proved that the image mosaicing algorithm inthis paper has higher performance than other algorithm.3.The VIBE algorithm is used in target detection, which are basis fortarget tracking. Then we propose an anti-occlusion target tracking algorithmbased on Kalman and TLD. The experimental results show that the algorithmalso has strong robustness and effectiveness even when the target is blocked.4. An integrated software system is designed, including image mosaicingalgorithm evaluation module，target detection and tracking module, integrate module. Finally two target tracking examples based on mosaicing video aredemonstrated.