The Research And Implementation Of Target Tracking Algorithm Based On Monocular Vision

With the fast development of information technology, The application of computer technology can be seen in every corner of our life. As an important computer vision technology, target tracking exhibited increasingly pivotal role. More and more scholars have paid attention to the research of target tracking technology. This thesis will discuss in detail the target tracking technology, and study a target tracking algorithm with good performance. the following is my main work in this thesis.First, this thesis study a multiscale retinex algorithm to deal with the problem of illumination change. The algorithm extract a reflected light image that related with target texture information. And the reflected light image has the illumination invariant feature. By using the reflected light image in the target tracking algorithm, the effect of the illumination change on the tracking results greatly reduced.Second, this thesis establish an affine transformation model to deal with the problem of target deformation. This model can correct the target image of deformation, and reduce the effect on the tracking results that the target deformation bring to.Third, this thesis establish a local appearance model of target image to deal with the target occlusion problem by extracting the HSV color features of target image block. This model can effectively determine the blocked area of the target image, and reduce the effect caused by occlusion.Fourth, this thesis establish a joint model based on the bayesian theory framework. The model contains the local model based on image block color features and the global model based on sparse representation. We can get the likelihood of each candidate target image by the joint model. And the candidate target image that has the highest likelihood is the tracking results in the current frame.Fourth, this thesis have carried out experimental analysis and proved the validity of the algorithm for each algorithm. Especially in the final comprehensive experiment, this thesis compared the algorithm to the other popular algorithms and effectiveness of the algorithm is verified.