Experiments with Photogrammetry
Photogrammetry is the means by which a 3D model can be constructed using photos as the sole input. It’s used extensively in archaeology to document artifacts in such a way that it allows specimens and sites to be explored in great detail without fear of having the original crumble into dust upon touch. I recently made a post about the visual fidelity of a game called the Vanishing of Ethan Carter, the game is beautiful and the studio responsible for it made extensive use of photogrammetry in their workflow.
The problem with using it in gaming is that a lot of editing is needed to make the art assets fit for purpose. The geometry needs to be optimised in order to ensure it remains playable on lower-end hardware and holes in the model, which are symptomatic of missing data, need filling in. It’s not quite as easy as just taking a bunch of photos and slapping it into a gaming engine.
I’ve decided to try my hand at it, using a souvenir I bought on holiday. See below for the subject matter and end results.
Everything depends on the quality of the source materials, if the photos are blurry or the lighting is inconsistent, then problems will begin to emerge. I’ve used the free version of Photoscan for this and I haven’t done much touching up of the above model but you can see that it’s somewhat faithful to the original subject. Areas that the camera couldn’t see due to occlusion remain missing in the 3D mesh data. See the images below for photos and screenshots of both the physical and authoring environment.
This is a rather small object so the camera really came through for me. The lighting conditions were ideal as well, there were no spotlights or inconsistent sunlight to worry about. The image below is of the authoring environment in Photoscan.
It’s an order of magnitude cheaper than 3D laser scanning but, if you don’t need extremely high fidelity reproductions, it’s perfect.