Playing with photogrammetry
Building on my prior research into this subject, I wanted to experiment more practically with this process…
Photogrammetry is the science and technology of obtaining reliable information about physical objects and the environment through the process of recording, measuring and interpreting photographic images and patterns of electromagnetic radiant imagery and other phenomena.
Wikipedia
For this I used an iPhone 11 Pro set to uncompressed RAW with manual controls (fixed aperture, white balance, shutter speed/exposure etc) so that all the images were of the highest quality and differ only in camera position. These settings are optimal for feeding into a photogrammetry app, in my case [Agisoft Metashape] 1.7.
For my first test I used two fluorescent desk lamps above and to the sides of the subject which was several items from my office desk. The subject was several items from my office arranged on my desk; a pig lighter on top of a wooden toy, on top of a portable speaker; various shapes, angles, textures and colours.
After using Metashape to delete some floating artefacts, the final results can be seen here:
The end results are very impressive to me for a first try, and have certainly opened up the possibility of using photogrammetry for further experiments and projects.
This could be especially useful when working with a team with no 3D modelling experts, as interesting models can be created with good photography captures as the source material. I think this process would really. shine when a project requires real spaces or objects to be turned into models.
The scan is quite accurate apart from some issues at the very back where it was difficult for me to hold the camera, so i am sure with better 360 degree access to the subject a seamless model can be produced.
One challenge of this process may be that it produces models with quite a large number of triangles and vertices, which may not be suitable for projects that need to run on limited hardware, such as phones, tablets or the web. My model has 441.4k triangles, and if I wanted to use this in something like Mozilla Hubs, would need to undergo a heavy reduction process to get it down to the 40-50k triangles range, at which point the model may be unrecognisable and unusable, so this must be considered.
Possible next steps
I’d like to try something larger and use a professional camera such as the Sony A7 R III which captures at very high resolution, and I think could provide a much more accurate scan, albeit with greater compute power (and therefore processing time) required. I am confident some truly breathtaking results could be had in this way.
Some uses of this process that spring to my mind include; creating archives of objects, replicating real spaces in digital worlds, a quick way to create aesthetically interesting 3D worlds with real-life atmospheres, as a way to capture street art within its immediate surroundings, and with the 3D texture of the objects its on.
