IMPROVED TRACKING USING P-N LEARNING

Abstract

The tracking and detection of target forms the basis of analysis and understanding of visual tracking.In adaptive tracking by detection framework,the object model is known.But due to fixed object model,the cluttered background may generate false positives or some object appearences cannot be detected.Thus a novel approach called Tracking –Learning – Detection(TLD) is used here which is composed of three subtask.The tracking block estimates frame to frame motion.The localisation of appearences observed during tracking is carried out in detection block.Here learning component estimate detector errors and update it to ignore these errors.Using above cascaded approach,computation time is reduced.The real time implementation of Tracking, Learning and Detection is explained and implemented on benchmark sequences.Thus significant improvement is achieved over state of art methods.