Difference between revisions of "Photogrammetry"

Latest revision as of 15:38, 29 June 2021

Origin

Photogrammetry is older than one might think. The name goes back to a preussian architect Albrecht Meydenbauer. First called Photometrographie. It's said that the name comes from photos= light ; gramma= which is drawn; metron = measure. But that was not the first time the idea came up. In the year 1849 A. Laussedaut thought about how photos could be used to create elevation maps. But between his version and today's photogrammetry there is a huge difference. In its early days photogrammetry was mostly used for creating DTM (Digital Terrain Model) for a military purpose. Eventually it emerged into geography and was used for GIS. Photogrammetry went through a process from analog to numerical to analytical to digital.

How it works

In the most basic sense you determine the position of photos in a 3-dimensional room. The overlap of the photographs is where the geometry is created. For the best result you need to create a stereo image. A stereo image is described as a photo that has two angles from a single focus point. It works like these images you probably have seen as a kid, where you need these red/cyan glasses to see them correctly. it's essentially that, the difference is that the computer defines the coordinates while you do not. While there are other methods this is the way most photogrammetry software works, when you use photos. While today algorithms can produce good results without this practice, it's more likely to get good results if you use this method.

Surfaces created from photogrammetry can either be:

• point based
• triangle based
• grid based (mostly landscape)
• or a hybrid

Because photogrammetry delivers too much information there needs to be an approximation. In this case it is interpolation. It reduces redundant data. We can differentiate, in the most basic form, between. Linear Interpolation: surface defined from 3 points Bilinear Interpolation: surface defined from 4 points

Other issues with photogrammetry are: Doesn't work very well with translucent Object, doesn't work well with to thin surfaces and has too much trinagles for some use cases

Other variations of photogrammetry

Lidar

Lidar stands for Light Detection and Ranging. Lidar sends electromagnetic energy(infrared) and records the scattering that is happening(it works similar to radar). It can achieve a more accurate reproduction of an environment, but that doesn’t mean it's the golden egg. Most of the time classic photogrammetry achieves the same results and it is cheaper.

Radar

Radar stands for Radio Detection and Ranging. It sends out radio magnetic energy. It is normally used to determine the distance , angle, or (velocity) of objects. While it can produce fairly good results it is limited by its range, susceptibility to interference(noise, clutter, jamming) and the problem with the doppler effect.

GPS

GPS stands for Global Positioning System. It may seem odd. What does GPS have to do with photogrammetry? If you have three points you can triangulate(determine) your position. If you now create different photographs you can align them accordingly to their position. With this method you can create a terrain-map. And because GPS also gets the height you now have a 3d Map.

Tools

Free

• AliceVision Meshroom
• Regard3d
• OpenMVG
• Colmap

Paid

• Agisoft Metascape
• RealityCapture
• Autodesk Recap
• Photomodeler

Further Reading

Photogrammetry Thread, best practice Polycount Forum thread.