1. UKM Learning and Research Repository
2. Master Thesis / Tesis Master
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	Incorporating CAMshift algorithm and support vector machine for object tracking
Authors:	Nur Ariffin Mohd Zin (P56135)
Supervisor:	Siti Norul Huda Sheikh Abdullah, Associate Professor Dr
Keywords:	CAMshift algorithm Support vector machine Object tracking Computer vision
Issue Date:	18-Jan-2013
Description:	Object tracking is one of computer vision tasks which has rapidly gained a lot of interests among researchers. In accordance with this trend, a study concerns on object tracking by adopting CAMshift algorithm is proposed. CAMshift, which is a derivation of Mean-shift algorithm, refers on target object’s colour probability distribution to locate the possible target object’s location in the subsequent frame. It is simple yet robust to track efficiently even during partial occlusion. A possible problem in CAMshift algorithm is that the mean of probability distribution becomes inaccurate when one or more foreign objects that share the same features with the target object are very close to one another, resulting these objects are in the same search window. Therefore, this study proposed the employment of machine learning technique to verify the back-projected representation of the target object and to generalize between target and non-target object. The aim is to track the movement of the target object to wherever it goes with the presence of ubiquitous objects, multiple backgrounds and different angles while maintaining the search window is covering the right target object. According to the proposed method, the target object’s back-projected images were trained and learned using Support Vector Machine (SVM). An orange coloured ball was used as the sample target object, which captured using a video camera running at 30 frames per second. About 1799 samples of ball’s back-projected images were used to train the SVM classifiers. The proposed method was tested to track the ball in three different environments; easy, adjacent and cluttered. Comparisons were made with the classical CAMshift in terms of tracking accuracy. Results have shown that the fusion of classical CAMshift and proposed method yields more robust and accurate tracking mechanism with up to 80% of accuracy.,Master/Sarjana
Pages:	97
Call Number:	TA1634.N847 2013 3
Publisher:	UKM, Bangi
URI:	https://ptsldigital.ukm.my/jspui/handle/123456789/476165
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.