AbstractRegistering shapes is fundamental to many tasks in computing such as fingerprint recognition and motion and image analysis and detection.
Particularly for three dimensional (3-D) shapes, the advantages of visualising data in 3-D such as the ability to represent more details enable improved analysis due to this representation.
For shape matching, the iterative closest point algorithm (ICP) has been the dominant algorithm for such purposes. ICP, however, has several limitations such as its expensive computational cost and its need for a good initial transformation.
Molecules have been represented as 3D point clouds in several tasks such as drug discovery and exploration of the functions of proteins by allowing the 3D point cloud to overlap another molecule fully or partially. Exploring such shapes matching in virtual reality (VR) also provides an immersive visualisation and analysis experience and a way to easily use more intuitive gestures using hand controllers instead of a mouse and keyboard to interact with the shape matching process. This allows a faster-repeated test of the matching as required and enhances the efficient exploration of the 3-D shape.
This thesis aims to develop and evaluate shape matching algorithms based on the standard ICP algorithm and a virtual reality visualisation application for visualising and interacting with the 3-D molecular protein shapes. It proposes three ICP variants developed to improve the registration of 3-D molecular structures. The approaches used in the variants are k-means clustering for partitioning the search space for correspondence, the use of metadata information to enhance meaningful matching and reduce the search space for correspondence, and the use of partial matches to register a subset of the shape and reduce the search space.
The algorithms are evaluated using 3-D molecules under different conditions such as noise levels and mutations to compare their computational speed, convergence properties and the quality of the matches for the 3-D molecular structures.
Molecular structures form the basis of all organisms. Visualisation of molecular structures is important because of its applications such as in drug discovery. This research further presents a demonstration of an exploratory immersive virtual reality application developed in Unity3D for visualising the matching of 3-D molecular structures using the HTC VIVE VR system and the ICP algorithms.
|Date of Award||27 Apr 2022|
|Supervisor||YONGHUAI LIU (Supervisor) & ARDHENDU BEHERA (Supervisor)|
- 3D Shape Registration
- 3D Shape Matching
- Iterative Closest Point