Data selection method and apparatus in visual tracking system based on incremental learning

摘要

PURPOSE: A data selection method in an object tracking system based on gradual learning is provided to effectively learn the outer appearance model of an object. CONSTITUTION: A kind information assigning unit(321) presently assigns kind information to tracking data and comparison tracking data of the update stage. A probability calculation unit(322) calculates the probability of the tracking data and the kind information tracking data located in the local domain of the comparison tracking data. An entropy calculating unit(323) calculates the entropy in the local domain of the tracking data and the comparison tracking data. A data sorter(324) classifies the tracking data into kind boundary data and kind non-boundary data by using the calculated entropy.