Scene-based Localisation

01 Mar 2014 Information Systems Technology and Design Image Processing, Computer Vision

Ngai-Man (Man) Cheung

Navigation based on GPS alone is not reliable for military operation. GPS jamming devices can be assembled easily with rudimentary engineering components and skills. Also GPS localization error is inevitable in urban canyons with dense collection of high buildings. The project aims to determine the location of the camera when the query image is taken without GPS assistance.

The team proposes to use the visual fingerprint of the scene environment to query a geo-referenced building database together with camera pose / location recovery algorithms. The project targets city-scale recognition, which deals with several orders-of-magnitude wider coverage and is very challenging. Together with camera pose / location recovery, this enables reliable localization at the city scale.

The team plans to research new mobile building recognition algorithms and systems to advance the state-of-the-art.  In particular, research algorithms that are robust for dynamic wartime situations such as partial destruction of buildings, and breakthrough technologies and system designs that enable secure and reliable localization entirely on the mobile device.