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

Title:	An accelero-visual markerless hybrid tracking technique for mobile augmented reality
Authors:	Waqas Khalid Obeidy (P63419)
Supervisor:	Haslina Arshad, Prof. Madya Dr.
Keywords:	Accelero-visual Markerless hybrid Tracking technique Mobile augmented reality Universiti Kebangsaan Malaysia -- Dissertations
Issue Date:	13-Oct-2014
Description:	Augmented Reality (AR) technology is perceived to be evolved from the bases of Virtual Reality (VR) technology. Where VR is a fully computer generated virtual content, Augmented Reality is the combination of the real world and the virtual content. The direction of research in the field of AR has been shifted from traditional HMD and bulky Desktop based mediums to the mobile devices such as the PDA’s and smartphones. These devices are equipped with display systems, cameras, graphics, GPS compasses and accelerometers which are enough to serve the basic requirements of the AR. Since AR demands seamless detection of objects in real-time, it faces major obstacles in the area of tracking which assures its real-time performance capabilities. Continuous tracking in AR applications is essential for registering and augmenting the digital content on top of the real world. However, tracking on handheld devices such as PDAs or mobile phones enforces many restrictions and challenges in the form of efficiency and robustness which are the general performance measures of tracking. Popular visual fiducial markers perform well but limit the interactivity and are constrained to a range of photos or objects encapsulated within a specific border. Markerless tracking offers solution to the limitations of marker based AR but is computationally very expensive, especially when implemented on low end devices such as smartphones which hinders its real time performance. Hybrid tracking techniques overcome these limitations by combining markerless tracking with sensors available in the modern day smartphones. This work focuses on the development of an Accelero-Visual Markerless Hybrid Tracking Technique which combines the visual feature based tracking with the accelerometer sensor of the smartphones in order to make the process of tracking more efficient and robust. Analysis is performed for the visual and sensor based tracking approaches required to design the hybrid tracking technique. For visual tracking, the best keypoint detector and descriptors are analysed. Different kinds of texture based images acquired from the metaio dataset are used for these tests. The best choice among all the tested detectors is ORB which is fast as well as scale and rotation invariance. FREAK’s description and matching performance of less than 8ms beats its competitors by far. A real-time visual tracking technique has been developed based on ORB detector and FREAK descriptor. Accelerometer can provide the shake and tilting values and can help speed up the tracking process. To improve the performance of visual tracking technique, this research combines the values acquired by the accelerometer with visual tracking for the development of an Accelero-Visual Markerless Hybrid Tracking Technique. This technique is implemented on Samsung GALAXY Note smartphone which fulfils all the basic requirements for successful implementation of AR on mobile platforms. The results of the conducted efficiency and the robustness tests on the benchmark tracking datasets such as Metaio, Vienna and Mikolajczyk proved the tracking performance has been significantly improved after the implementation of the proposed Accelero-Visual Markerless Hybrid Tracking Technique. Siginificance tests conducted for evaluation of the technique showed that it could produce very efficient results of tracking at 49 Hz, at least 19Hz faster frame rates than previous researches. Moreover various benchmark robustness tests on different datasets showed great results over all the tested sequences specially overcoming the limitations of scale and rotation invariances found in previous works.,Master/Sarjana
Pages:	170
Publisher:	UKM, Bangi
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.