In order to solve the problem of lack of discriminability in locality -constrained linear coding (LLC) and collaborative representation (CR),an algorithm for collaborative representation tracking is proposed based on multi-kernel fusion and locality constraint.Via LLC,firstly,the sample’s lo cal structures are introduced into CR to achieve the better performance on image classification.Secondly,CR is extended into high dimension multi-kernel fusion space using the kernel function so as to enhance the discriminability of target feature in dictionary template and CR.Finally,the candidates′ representation is put into the classifier and the candidate with the biggest Support Vector Machines score is regarded as the target.And the experimental results dem onstrate that the proposed algorithm is able to maintain an accurate and robust tracking when the target appearance unde rgoes huge variations.