3D photogrammetry: turning photographs into 3D models

3D photogrammetry is the process of reproducing a physical item in order to make an accurate 3D model. It can be used to capture everything from people and historical artefacts to aircraft, buildings and even larger structures.

"It's a very good way of making an object tangible and immersive, so you can look around it and see the texture of it," says John Maurice, European Product Marketing Manager at Canon Europe. "You can also do this in a CG render, but it's not the same because you never get all the nuances that photogrammetry captures."

The applications of digital photogrammetry and 3D modelling are endless, with the technique employed as a solution in manufacturing, engineering, design, entertainment and healthcare. "Creating 3D models of somebody can help with a diagnosis and the monitoring of treatment," explains John. "And if you work in industry, being able to make a 3D model cuts down on the expensive prototyping stage."

3D photogrammetry is a very specialised field and the process is a meticulous one. "You need to take lots of pictures of the subject, and typically there needs to be a two-thirds overlap between the images in order to create the 3D map," John says. "So, if you're taking the shots with a single camera, you need to move it slightly but very precisely, so that you get that overlap."

Choosing affordable cameras is a priority when you're looking to populate a 3D photogrammetry rig with 100-plus bodies. "Entry-level DSLRs are good for this type of work," says John. "And affordable lenses, too. 50mm lenses are popular, because they're cost effective and good quality."

"Part of the reason why we haven't moved to full-frame is that the depth of field is greater with all the APS-C cameras, and that's really critical to the processing of the data."

In its full-body rig, Sample & Hold uses a mix of 100mm, 85mm and 50mm lenses such as the Canon EF 85mm f/1.8 USM and the Canon EF 50mm f/1.8 STM, with the longer focal lengths used for picking out important areas such as the feet and the hands. The head rig uses a combination of 40mm and 50mm lenses, such as the Canon EF 40mm f/2.8 STM.

"We'll always stick with prime lenses," says Sam, "partly because there's no worry that things are going to shift, especially if we're moving the rig.

"Entry-level cameras aren't designed to do what we're doing with them, so there can be discrepancies in the shutter timings," he continues. "To get around this, we use flash to control the exposures. We work in a dark room and run our cameras at 1/3 or 1/5 sec shutter speed, and the flashes all fire simultaneously to expose the images. We control the cameras using Smart Shooter software, and sync everything using Esper TriggerBoxes."

Esper provides multi-camera and 3D scanning solutions to service providers in the video game and visual effects industries, as well as for advertising, academic research, bullet-time photography booths and beyond. Company director Nick Foots says that the most important thing to consider when setting up a 3D photogrammetry system is the final output.

"It's the first question I ask anybody during a consultancy call: what is your end use? As in, what resolution and format? And then you can work back through the whole pipeline based on that," Nick explains. "So, if something's only ever going to be used for e-commerce on the web at a not particularly high resolution, then, dare I say it, there's no point in having 60 cameras pointing at the thing.

"If you're going to do a 4K video game, then you may need upwards of 160 cameras, but if you're going to create 'mini-me' 3D models of people using an FDM printer, then you're really only going to need 30 to 50 cameras. There's just no way a 3D printer is going to replicate the resolution that you're going to get from using any more than that.

"If an object's small enough and it doesn't move, you can probably do it quite economically with three or four cameras on an arc, a bit of controlled lighting and a turntable that you move by hand," Nick continues. "You can also do perfectly good photogrammetry scans with a single camera, although you'll need to consider if that's scalable for what you want to do, and how it's going to fit into the whole workflow."

Starting with a single camera is something that Sam at Sample & Hold recommends. "Anyone can have a go at photogrammetry out in the field, using daylight and a single camera," he says. "With a subject that's not going to change or move, such as a house, you don't need to use multiple cameras, you just use one camera from multiple positions.

"There are lots of tutorials on YouTube that show you how to do it," Sam concludes. "And then you can run the images through something like Meshroom, which is a free, open source bit of software, and produce 3D models. You can create some really cool stuff this way."