What is Photogrammetry?
Photogrammetry works by taking a large number of photos of an object from many different angles, processing them with computer vision, and building a textured 3D model of the subject matter.
Results will vary, but the process works best when there are a large number of images (generally hundreds) showing every part of the object from a number of different angles, with a significant amount of overlap between the images.
In order to get the best possible results, it is important that:
- The subject stay in the same position, in the same lighting for every photo that is taken.
- You light the object as well as possible, but ensure the lighting stays consistent throughout the photo taking process.
- You don’t use flash, or a handheld or camera light, since this will make the lighting different for every image and seriously interfere with the software matching up areas.
NOTE: Your resulting 3D model will appear exactly the same as in your photos. Remember to ‘tidy up’ and remove any unwanted cables, cloths, debris etc. from your subject model.
To prepare your content, you need a good camera. Check out the tips below:
- Any modern smartphone will provide good quality pictures at default settings, usually these will be around 2 MP, which is sufficient, although higher resolution is always better, since you can reduce it if needed, while lower resolution can not be improved.
- Ensure the photos are crisp, with no blurring due to motion or ‘depth of field’ effects such as portrait mode. Again, default settings should be fine (If using an iPhone, be sure to turn off ‘live photo’, since it also captures a short video sequence.
- Wide angle views should never be used, since they will distort the shapes around the outside edges of the photo, preventing the software from matching the features.
- The goal is to capture as many angles as possible of every portion of your subject material, with particular attention to any areas which are more complicated (i.e. areas with nooks and crannies).
- It is NOT important to take the photos in a particular order, or to be consistent in the pattern you use to capture the images. In fact, adding a selection of random images at the end of the process is very much a good idea to ensure the ‘rules’ you followed in taking the pictures didn’t accidentally exclude some helpful views.
- It is helpful to establish some sort of pattern however just to make sure you capture the object in its entirety.
In the example below, an object which is roughly cylindrical and 5 feet or so tall is shown. In reality the object has a great deal of surface detail which is desired in the resulting 3D model.
One way to approach this is to capture the object in a series of circular ‘passes’. Each pass is a sequence of photos taken at multiple surrounding views, followed by a similar pass further away. This process is repeated at multiple heights, because of the height of the object, ensuring that there is overlap at the top and bottom of these ‘layered’ views. A shorter object would require fewer ‘layers’.
Again, it’s important to note that being rigid in this pattern is not necessary, other than to ensure nothing is missed. Adding additional ‘random’ views throughout will add more information to the process. While the result will generally be hundreds of images, they can usually be taken within 15-30 minutes with little trouble. Be sure to take more pictures surrounding ‘difficult’ spots such as cutouts and projections to ‘help’ the software understand what’s happening with the structure.