Fast, efficient method for reconstructing 3D spacial data online from regular photos
A research group at Tokyo Institute of Technology is developing a system that quickly creates 3D spacial data from photos taken with a digital camera.
"We take pictures using an ordinary digital camera. This camera has an Eye-Fi card, which sends the pictures to a PC wirelessly. This system uses the received pictures to rapidly create a 3D model. A feature of this system is that it doesn't use any information other than the pictures to construct the 3D data."
"Other people are researching similar systems. What we're specializing in is the online aspect. In other words, what's interesting is, you see the results as soon as you've taken the pictures. For example, when aerial photos are processed after saving them, if you need to take more pictures, you have to go out again. But with our system, you can see on the spot whether 3D measurement was successful, so if you don't have enough pictures, you can just take some more."
This system uses SfM, or Structure from Motion, which estimates the 3D shape and camera position from several pictures of the same scene. To reproduce spatial position data, it repeatedly identifies and matches characteristic points between two pictures.
The group has achieved an efficient 3D restoration method, by detecting pictures that aren't suitable and doing overall optimization in line with the cumulative error.
"This system doesn't respond to featureless surfaces well, but it can handle most scenes with features that show up in pictures."
"There's often a need to render or create surfaces. The system we're presenting here is still at the initial stage; we'd like to develop it further, so it can handle surfaces online, rather than just points."
"Taking pictures without any constraints and calculating 3D information is a challenging project in itself. Anyway, we have this basic technology for obtaining 3D data, so we'd like to get everyone's feedback on possible applications."