ImageModeler 2009

ImageModeler 2009 extracts 3D information from still images to construct accurate 3D models and scenes, to which you can apply realistic textures taken from the image.

It was originally developed by RealViz, and was acquired by Autodesk along with Stitcher and Movimento in May last year. However, it's still not available through Autodesk's normal sales channels, only through an online sales site.

When working with multiple images, you have to identify common characteristics on the object in each of the photos loaded in the viewer, placing calibration locators on the corresponding points. You then define a point of origin — usually a corner of a building — using the World Space tool in conjunction with these locators. A reference distance or measurement of two points in the scene is required for accurate scaling.

In the background, ImageModeler calibrates the parameters of the camera that took the shots (such as position, rotation, focal length, distortion) to allow it to calculate the 3D coordinates of the object.

Once enough points are matched, the software announces that the calibration has succeeded — or it should do. In our first attempt, the 3D space construction was imprecise. Fixing this involved tweaking the locator position to make it more accurate, but of greater help was adding constraints, such as setting the right-angled corner of a building, or specifying that a set of locators are on the same plane.

You can use distant constraints in this version, using the reference distance to aid calibration between two locators. You can also add 3D coordinates, if you have CAD data or GPS survey information.

It’s simpler when working with single images, as you just define angles and perspectives in the shot to gather the spatial coordinates. It’s a similar process for 360-degree panoramic images, which can be used to model complete interiors.

If all goes well, ImageModeler can retrieve the spatial coordinates of any 3D element from its projections and use the 3D points to build a polygonal mesh. This can then be manipulated – for example, you can use the new Bevel tool to add more details by subdividing an edge. Polygon primitives and imported 3D objects can also be added to scene and edited.

You then use a choice of planar, cubical, cylindrical or spherical mapping projection to build a UV map of parts of the object mesh, and ImageModeler will provide a selection of shots from which to extract textures.

This is simple with a single image or panorama, but with multiple images you can define the resolution of the texture and base the extraction on the best image for each part of the model. You can also use a Smart Blend mode to automatically smooth out the texture. You’re then ready to use the textures as materials, mapped accurately onto the models.