Visual tracking is one of the top topics in computer vision. Tracking of small and low-contrast infrared objects is also an attractive research field for its abroad application in military. In this thesis, we focus on object tracking and detection in infrared environment on the basis of the visual object tracking algorithms. The contribution of the thesis are : 1. By using the feature of the grayscale of object and factored sampling algorithm, we achieved the tracking of the fast-moving object in infrared scenes. When an object is moving in cluttered environments, tracking becomes a problem of nonlinear and non-Gaussian state estimation. In this situation, as a simulation-based method, factored sampling provides a convenient and attractive approach of computing the state posterior distribution. Experiments demonstrate that it is a robust tracker with factored sampling. 2. D. Comaniciu et al[5,32] have used the Mean Shift optimization method in tracking of non-rigid objects. We expand upon the ideas of [5,32], exploiting the useful properties of the feature of the intensity projection because the central symmetry of the 2-dimensional kernel function can not characterize object as theonly one. In addition, Gabor filters are applied here to enhance the contrast of theobject with the background. 3. Objection detection and re-tracking is also an important part for the trackingsystems. We use a simple way, i.e. background subtraction to detect the movingobjects. About object re-tracking, by using the object template, we search the mostprobable location where the candidate object has the most likelihood with the objecttemplate. And the Gauss pyramid is used here to enhance the exhaustive searchalgorithm.
修改评论