target tracking is the tracking of an object in an image sequence. target tracking in image sequence consists of two different parts: 1- moving target detection 2- tracking of moving target. in some of the tracking algorithms these two parts are combined as a single algorithm. the main goal in this thesis is to provide a new framework for effective tracking of different kinds of moving targets. using two-step algorithm and employing the extracted target features, especially the color we are able to achieve reasonable detection and tracking results. we have shown that by employing the local mask over the target which is an isotropic kernel and defining a similarity function between target in the present frame and the target-candidates in the next frame we are able to search effectively for the target in the next frame. in fact, instead of exhaustive search in the next frame we are employing a time-effective approach which estimates the target in the next frame. in the most cases the object tracking dont have isotropic perspectives. so, we introduce a new non-isotropic base on metric distance transform to make target and candidate models .the similarity measure of the target and the candidates in the next frame is done with the bhattacharyya distance. the bhattacharyya distance calculates the correlation between the target and the candidates. after location estimation of the target in the next frame a two-class classifier, in which the support vector machines were used in here, is used to find the exact boundary of the target. in this step a feature vector of the original target is formed and classifier is trained with this vector in a way that pixels belonging to target are in one class and the counterpart in the other class. next, in the coming frame using this classifier, we testify the pixels in an oval with the size of two greater than the oval including target obtained by the motion estimation model. with this approach the accumulated error of the first stage is compensated and exact boundary of the target is found. moreover after a while we can update the classifier and hence increase the accuracy of the tracking algorithm. we have shown in our experiments that the proposed algorithm is robust against the camera movement, slight target occlusion, complexity of the background and illumination changes.