Seminar Series 2/2009: "An Improved Stereo Framework"

Mr Ng Oon-Ee (MEngSc (resesarch) student)

Date: 2009-02-18
Time: 15:00 to 16:00
Venue: 9-3-01


Computational stereo vision processes a pair of slightly-differing views of a scene to produce a depth map of the mutually visible surfaces. Though the stereo matching problem is inherently ambiguous, many algorithms have been developed to solve it. Current stereo algorithms are primarily classified based on a local/global dichotomy which does not provide a comprehensive description of the entire computational stereo vision problem. A new framework is proposed which describes the stereo vision problem as modular in nature. Some current stereo algorithms are classified according to this framework to demonstrate its utility. The new framework is expected to spur novel techniques to solving the stereo matching problem in a modular manner, rather than in terms of entire novel algorithms. Some proposals on possible novel modules are proposed and described in detail, along with suggested methods for quantitatively measuring the quality and efficiency of the results.

About the Speaker

Mr Ng Oon-Ee is our current 1st year HDR (MEngSc (research)) student under Dr V. Ganapathy’s supervision. He will be presenting his research work during this HDR seminar as a fulfillment of candidature conversion from MEngSc (research) to PhD.

There will be a Q&A session after the 20 minutes of presentation. Please come and support!