Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.582426
Title: Continuous 3D hand tracking as planar target
Author: Morshidi, Malik A.
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 2013
Availability of Full Text:
Access through EThOS:
Full text unavailable from EThOS. Restricted access.
Access through Institution:
Abstract:
This thesis focuses on the development of an articulated 3D hand- tracking framework where a single-view technique using a pre-calibrated low cost webcam is explored. The tracking method is based on a novel extension from the standard particle filter by introducing gravitational force as a new attribute in addition to every particle’s weight. The par- ticles close to the current observation are attracted closer to provide an improved posterior distribution that results in fewer particles utili- sation as well as better tracking accuracy. Experiments show the pro- posed method outperforms the standard particle filter and its variants, as well as Camshift guided particle filter. The hand features extraction utilises convex hull and convexity defects methods to extract dominant features from hand silhouette. Using the hand geometry knowledge, methods for labelling finger and determining left or right hand, as well as addressing the finger articulation problem are proposed. Com- bining the tracking method and the hand features extraction has the advantage in better occlusion handling and tracking. The pose of an object is estimated using the robust planar pose method that is incor- porated in the Augmented Reality Tool Kit, which is a marker-based computer vision tracking library for solving the camera projection ma- trix. The ambiguity in markerless hand pose estimation and tracking is solved by selecting a better combination of hand feature points as control points. An OpenGL 3D model featuring finger kinematic and articulation constraints is developed using a combination of small el- lipsoids that correspond to every rigid part of a hand. The model is augmented back at the estimated hand location in the original video sequence to produce a seamless see-through effect. Finger articulation is addressed by introducing a dynamic hand model points selection that can be used interchangeably while tracking. These combinations of different model points ensure continuous tracking of 3D hand as planar target.
Supervisor: Not available Sponsor: International Islamic University Malaysia ; Kementerian Pengajian Tinggi (Ministry of Higher Education ; Malaysia)
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.582426  DOI: Not available
Keywords: TA Engineering (General). Civil engineering (General)
Share: