Object Detecting And Tracking For Space Moving Image Sequences

Moving object detecting and tracking is a major search area in computing vision. While the research work progressed, it meets much more and much tricky problems. Also need to deal with many kinds of applying conditions.While facing the applying situation of unstable cameral and un-continuing object tracking, this paper proposed kinds of solutions to solving these problems and expand the applying field of object tracking. By doing registration of before and after frames we can use the algorithms which were fit for the stable back ground video such as meanshift and particle filter to solving the unstable cameral videos. The frames before and after registration process is also applied to illumination compensation and feature point calculation, image processing method, so that the final deal was more accurate.In object detecting this paper proposed two different solutions for adaptively object detect for different data scenario. The first solution is a progressed Gaussian background model way to achieve the target detection function, the second solution is meanshift adaptively object detection algorithm based on object template matching.This paper also proposed a solution that combined kalman filter and NCC template matching algorithm to dealing with the continuing tracking problem. When the moving object tend to be static and the kalman filter cannot tracking it any more the proposed algorithm can take in charge and retain the object, and it can achieve the demand of continuing object tracking. The following are finished works of this paper:(1) Before and after frames registration based on illumination compensation Proposed a before and after frames registration algorithm based on stable feature points. And the stabilities of the feature points are matters a lot to the final results of the before and after frames registration. By distracting stable feature points from about10adjacent frames we can simplify the feature point distract work and have the same effect. By using less stable feature points to calculate the registration parameters then we can obtain the accuracy registration results.(2) Adaptively object detecting algorithmThis paper proposed two different of adaptively object detecting algorithms to solving the adaptive object detecting problem. The first one is an improved Gaussian background model algorithm. The second one is the meanshift object detecting algorithm that based on object template matching. Both these algorithm can meet the demand of adaptively object detecting.(3) Continuing object tracking for moving images sequencesThis paper proposed a center point based NCC template matching algorithm (CPNCC), and combined it with the kalman algorithm for multi-object tracking. These methods can solving the uncontinuing object tracking problem. When the moving object don’t moving any more this method can switch to the center point based NCC template matching algorithm and keep the object in tracking.(4) The realization for the verification platform of object detecting of moving image sequencesAt the end we designed a verification platform to realize the three searching contents above. And the platform is based on the matlab GUI programming and separated into three modules. Each module is for one searching aspect such as before and after frame registration, and adaptively moving object detecting. At last we tested the platform to prove its effectiveness and integrity.