Guide: Photogrammetry
Photogrammetry is the science of making measurements from photographs. Using photogrammetry, we can create 3D models (of things, places, or people) from photographs, and embed them in stories. The Open Lab can work with BuzzFeed reporters to use photogrammetry to create immersive visualizations in your reporting.
There are two basic variations that we can help you create:
- 3D rendering of an object
- 3D rendering of an environment or room
In addition, we’d love to work with a reporter to create a much more immersive experience in virtual reality so that a reader can encounter clips from interviews, excerpts from the article and other audio assets as the explore the environment in VR.
Ben Kreimer ([email protected] or @ben.kreimer) and Ainsley Sutherland ([email protected] or @ainsley) can work with you to build out 3D models and immersive environments for your stories. You can find the lab fellows in #openlab-chat anytime.
| Rendering of an object: | Rendering of a room: |
|---|---|
| https://skfb.ly/GNrr | https://skfb.ly/NvMU |
 |
 |
The Open Lab fellows need some lead time to be able to assembly your photos into a 3D model, so definitely reach out if you’re thinking about introducing a model to a story. Some ideas for things that you could reconstruct in a model:
A sculpture, a famous car, a rare mushroom, a model testing out a new line of clothes, Jeremy Bentham’s Severed Head.
A famous writer's study, a crime scene, a jail cell, a cave, the living room of a family devastated by Chicago's homicide epidemic, the inside of a Pulse nightclub or the site of a horrific prison bus crash, or any place where giving readers a chance to look around would add to the story.
We've been using a service called SketchFab to host 3D renderings -- their HTML5 viewer works very well on mobile and browser, and you can embed a SketchFab rendering in the CMS.
Anything that moves while you are photographing the scene or object will be distorted or may disappear from reconstruction altogether.
Some surfaces are really hard to preserve in a 3D reconstruction. White walls, reflective surfaces, and windows will disappear. If you really need to photograph such surfaces in a reconstruction get in touch with Ben or Ainsley and we can help you figure out a strategy for getting images that will work.
Whether you're photographing an object or an environment, the basic principles are the same: you want to get a lot of photos with a lot of overlap, from a variety of distances and angles. You need both movement and overlap to produce a rendering. Skip down if you want to see a diagram of the photo process.
Movement
What you need to do is move around the room so that your body isn't in the same place each time you take a new picture. You should take a step forward or back to zoom in or out: using your camera's zoom function will reduce the quality of the rendering. So don't zoom, move your body to take pictures from different depths. The variations in depth and angle and camera position are what allow our 3D rendering engines to generate a profile of the physical space or the object.
You might have some experience taking photos for a panorama: that is exactly the wrong thing to do for photogrammetry. Definitely do not stand in one place and take a bunch of pictures without moving around. You have to move.
Overlap
The images in your photos should overlap a lot. You want 60-70% overlap. It will feel like a lot while you’re taking the pictures. You are going to take a lot of photos. Way more than you think you need.
It is sometimes possible to create a good reconstruction with fewer photos. Haley Campbell had a chance to visit with Jeremy Bentham’s Mummified Head last fall she took 17 photos, very few of them from the back or above. The head would have been a great subject for a 3D reconstruction so we tried taking the photos that she did take and feeding them to Autodesk Memento. The resulting rendering is pretty intense, especially if you zoom in. Note, the detail is very intense when you zoom in. Not everyone actually wants to see that: https://skfb.ly/OoHQ
With a more photos from the top and rear of the head, we could make a complete rendering.
Sample of photos you need to accurately reconstruct an object:
Diagram: Ben Kreimer
The image above was generated by Agisoft Photoscan in the course of producing a 3d model of a sculpture. Each blue rectangle represents camera positions during the image process. Each rectangle has a black line sticking out from it showing the angle of the photo in relation to the object.
Physically move the camera closer or further back instead of zooming in and out on the camera. Don't move any objects during the photography process, or else it may appear distorted or disappear in the reconstruction. Make sure each photograph has 60%-70% overlap with the images taken before and after it.
The stars and arrows in the image below represent the direction and placement of the camera throughout the space. The photographer usually stood with the nearest wall to her back, photographing towards the opposite walls and surfaces, as indicated by the arrows emanating from the stars. The notable exception is found at the bottom right corner of the screen, where she took photos inside of her closet, then rotated 90-degrees and began taking photos of the rest of the room.
We can assemble a reconstruction for you, but if you’re excited about trying to do the whole thing yourself, Autodesk Memento is cloud based software that will process a reconstruction for you. It is free to use. Memento allows a maximum of 250 photographs, however, so if you have more than 250 images you’ll need access to Agisoft Photoscan. We can help with that.
Sketchfab is currently the quickest and easiest way to to share your 3D reconstruction. It’s the YouTube for 3D content, allowing users to upload and embed their reconstructions into the BuzzFeed CMS, Facebook News Feeds, or any webpage, as easily as a YouTube video.
If you want to practice photogrammetry entirely within your phone, download 123DCatch -- it’s available for Android or iOS.
https://docs.google.com/document/d/18lCjVDlGWAXWqQkQ33DcYCc5Z0Th5maoWAA_HLjrwvs/edit
https://docs.google.com/document/d/1RQTYmHissJtG9euQ9Cr9brDkh39WlkRTy5bnqwwPNUc/edit