Online Failure Detection and Correction for CAMShift Tracking Algorithm

Tracking failure is an inevitable problem in any objecttracking algorithm. Online evaluation of a tracking algorithm todetect and correct failures is therefore an important task in anyobject tracking system. In this paper we propose an earlytracking failure detection procedure for the ContinuouslyAdaptive Mean-Shift(CAMShift) tracking algorithm. We alsopropose an algorithm to modify the tracker in order to correctthe detected failures. CAMShift is a light-weight trackingalgorithm first developed based on mean-shift to track humanface as a component in a perceptual user interface, but it easilyfails in tracking targets in more complex situations likesurveillance applications. With our proposed failure detectionand correction algorithm, CAMShift shows promising results inthe test video sequences

کلیدواژه ها:

CAMShift ، Tracking evaluation ، Failure detection ، Failure correction ، Moving object detection