3D models from low-end digital images

Share this on social media:

A new technique for vision processing could allow 3D models of buildings to be recreated from ordinary, 2D digital photographs. Current techniques for 3D imaging typically require expensive equipment and special lighting, whereas the new software could be used with most low-end cameras.

The software could be used to provide a cost-effective solution for museums and sights of national heritage. It could also provide accurate 3D reconstructions of archaeological artefacts, on site.

To function, the software needs photographic coverage of every part of the object it is being used to reconstruct. This object could be of any size, although larger buildings would need more images to obtain the correct amount of detail.

Once this has been collected, the processing occurs in two stages. Firstly, it determines the outlines of objects in the images, and separates the object from its background. From this, it tries to match different specific features (such as doors or windows) from the different photos, and estimates the cameras settings and the movement of the photographer. It then uses this information to give a basic, crude model of the object.

After this first stage has been completed, the software then performs an advanced optimisation algorithm that readjusts the model to give a complete, photorealistic 3D image of the object. The software does this by working backwards from the crude model, to see how it would need to be altered to give the original photos.

‘That’s tricky – it’s where the power of the method is,’ says Martin Weber, the CEO of Cambridge InnoVision, which has created the software, told imveurope.com. ‘It’s a bit like the initial model is made of clay, and then deformed to improve the agreement with the images.’ All the processing is completely automatic.

In addition to applications in the heritage industry, he also believes it could be used by utility companies, to model their unmanned stations and provide 3D maps for visiting engineers.

At the moment, the company is providing this as a service, but they hope that one day it could be supplied as a package with digital cameras for the consumer market. The software still has a few bugs when dealing with reflective surfaces such as glass or mirrors, and it hasn’t yet got a user-friendly GUI interface.