Undergraduate Projects (1999-2000)

3D Model Acquisition by Tracking
Matt Brown
The aim of this project is to provide a point and click means of building a simple 3D model of an object, by tracking lines in video images of the object. This could be used for generating objects for computer graphics, or reconstructing 3D models from video sequences.
Supervisors: Tom Drummond, Roberto Cipolla

Enhancement of PhotoBuilder
Rich Fisher
Improvements to the PhotoBuilder application are being made by the development of: (i) Colour compensation to correct for images with different light intensities, (ii) Re-building of occluded texture, and (iii) Correction of wide-angled lens distortion.
Supervisors: Duncan Robertson, Roberto Cipolla

3D Model Scanner Under Circular Motion
Alex Tan
Recently, a novel technique has been developed for the recovery of viewer motion which exploits the envelope of apparent contours (profiles) under circular motion. In combination with existing methods of surface reconstruction, this project seeks to design an automated circular motion scanner which can acquire the full 3D model of an object placed on a turn table in front of a fixed camera.
Supervisors: Kenneth Wong, Roberto Cipolla

Visual Guidance of a Mobile Robot
Rana Molana
This project looks at the application of 3D model-based tracking to a real world environment, to navigate a mobile robot mounted with a camera. Real-time tracking of an internal CAD model of the environment is used to determine the robot's position and guide it to various tasks.
Supervisors: Tom Drummond, Roberto Cipolla

Unsupervised Model Learning for Recogintion
Rob Fergus
Current recognition models need human assistance in choosing which features to use for recognition. In this project, statistical techniques are being developed and implemented to allow the computer to decide for itself which features are stable and distinctive of the class of object to be recognised, so automating the recognition process.
Supervisor: Roberto Cipolla

Automated Model Enhancement
Steven Woolston
A variety of Computer Vision techniques are being used to assist users of the PhotoBuilder application and automatically enhance the VRML models it creates. For example, feature detection and matching algorithms can be used to find correspondances of selected features between images, reducing the amount of input needed from the user.
Supervisors: Duncan Robertson, Roberto Cipolla