ImageModeler 9 (IM) is a powerful tool for generating photorealistic, image-mapped 3D models from 2D photography. It reproduces 3D space by calibrating corresponding 2D points between multiple photographs, a process called “photogrammetry.” Once the 3D space is defined, you can “trace” 2D objects within the photos using 3D modeling tools.
We got the idea that IM might also be used to accurately document built and existing conditions for construction. If so, such a tool would be an inexpensive, in-house alternative to laser scanning. However, after two weeks of tinkering, tutorials (Autodesk and third party) and field tests, we found that despite our best efforts this is simply not one of IM’s current capabilities. To be clear this article is about our attempt to leverage IM for a purpose it was not designed for and not a criticism of the product itself.
The crux of the problem is that BIM ready field documentation needs to have a certain level of reliable accuracy. IM was designed for visualization and, as such, produces minimally accurate geometry. As we field tested the limits of IM were apparent at every step, but we kept developing workarounds in the hope that we could bend IM to our will. IM, to its credit, stood strong.
First, simply capturing the right photos (photos that show complete edges of the building, from roughly the same distance and with minimal occlusions from surrounding elements) is a real challenge in the large scale world of buildings. Even if you manage to capture every edge of the building with the minimum of photos (calibration starts to become unmanageable after 12 photos) you will be unlikely to model detail elements from the same set of photos used for overall massing. (The work-around for this is to model details in separate files and combine them in another software; unfortunately, this effectively multiplies discrepancies of scale and alignment). Whatever quality the photos, calibration and modeling are ultimately limited to a screen resolution of 4096px X 72dpi that does not improve as one zooms in.
Modeling itself is another challenge. Objects are generated from manipulating subdivisions of primary solids – for those familiar with 3DSMax, this probably sounds fine. Unlike 3DSMax though, these tools are very basic and have limited snapping capability. Modeling of complex shapes is difficult and tedious. Modeling with reliable accuracy is practically impossible.
For field documentation, the model should be accurate to at least ¼”. IM can produce a model with accuracy down to maybe ½”. For larger the objects (say the 220’ long building I work in) the accuracy is more like 2”– 3”. Simply not good enough for construction.
The final limitation is exporting the model. IM models export as scale-less DWG’s with no origin. Thus, separate IM models must be aligned and scaled manually. IM to DWG geometry is NEVER orthogonal, though, which means Revit will not like the walls generated from an IM mass. Finally, even if you made it this far, geometry loss during import is common and difficult to prevent.
Generating accurate real world documentation is not IM was designed to do. If you have watched the marketing videos, what it does do is actually pretty awesome. If Autodesk is interested in taking this software in the field documentation direction, they will run into competition with PhotoModeler, which incorporates more sophisticated camera calibration and utilizes coded field targets. For visualization I am sure most find IM’s stripped down workflow perfectly adequate; especially for something that came free with your Autodesk subscription.
ImageModeler was presented to me last year as a cost-effective (albeit less accurate) alternative to laser scanning for document existing conditions. We started looking at it last year and found some limitations. Since I have read a few blog posts about it recently, I asked the modeler, Pierce Reynoldson, who tested it to share his thoughts: