TRACKINGMOTION-BLURREDTARGETSINVIDEO

Manyemerging applications require tracking targets in video. Mostexisting visual tracking methods donotwork wellwhenthetarget ismotion-blurred (especially duetofast motion), because theimperfectness ofthetarget's appearances invalidates theimagematching model(orthemeasurement model) intracking. Thispaperpresents anovel methodto track motion-blurred targets bytaking advantage oftheblurs without performing imagerestoration. Unlike theglobal blur induced bycamera motion, this paperisconcerned withthe local blurs that areduetotarget's motion. Thisisachal- lenging taskbecause theblurs needtobeidentified blindly. Theproposed methodaddresses this difficulty byintegrating signal processing andstatistical learning techniques. Thees- timated blurs areusedtoreduce thesearch range byproviding strong motion predictions andtolocalize thebest matchac- curately bymodifying themeasurement models. Index Terms-Tracking, image deblurring, pattern recog- nition, frequency domain analysis