Rule induction based object tracking

In this work, an object tracking method using a rule mining/induction technique is presented. Initially, a rule based classification algorithm is employed to classify the target frame into object and background. A sequential covering algorithm is used in order to extract the rules from the candidate frame. Extracted rules are then used for classifying the test samples obtained from the search region of the target frame. Classified test samples form the classification map which are used for calculating the new centroid to locate the object in the target frame. Temporal coherence (between frames) is maintained by updating the rule set during the rule extraction phase. Efficiency of the proposed method is established both qualitatively and quantitatively by comparing it with some of the state-of-the-art algorithms.