Project TACREA

This is a DSTA-funded project on localizing a user carrying/wearing a mobile camera in an urban area based on images captured from the camera. Such imagebased localization technology is particularly useful in unfamiliar high-rise urban areas, where GPS triangulation may not be adequately achieved due to line-ofsight obstruction. The prior data available is a 3D geometric model layout of the urban area, which may optionally include facade appearance data. The proposed

framework involves separate stages of recognizing buildings, determining image-to-model registration, and followed by pose computation.

            The problem involves partial observation (camera tells only part of what the user sees) and partial information (the models have only partial information, environment can occlude part of the structure by things like foliage..) But definitely not so partial location results J as it’s intended for urban and warlike scenarios where its critical to know where you are to metre range accuracy.

Many issues are to be tackled for solving this problem,

• Where to start looking and matching
• How to progress the matching process
• What is the most relevant and useful information from the model, secondary sources like maps
• How can these information sources be melded together to aid the search
• How to handle ambiguity in structural features
• Do avoid exponential database sizes
• How to use approximate models

How is the problem new ?

• It’s new to me :P
• It’s not structure from motion (stereo)
• The idea is to use as many sources of information together

Similar things happening at a more global scale Google 3D.

Will be more than happy for ideas and pointers…harishk@comp.nus.edu.sg