Photogrammetry R&D

[arve url=”″ maxwidth=”860″ /]

3D reconstruction of a burnt-out car. 1410 images shot with a Canon EOS 450. Processed with Reality Capture.

Photogrammetry is the process of extracting visual data from an array of photographs for the purpose of generating fully textured three dimensional representations of objects and environments.


Photogrammetry scan of a small scale toy with 515 corresponding aligned cameras (Screenshots)

It is used in different fields like cartography, archaeology, and cinema. Only recently this technology was adopted by the VR and video game industry, since it raises the visual quality of the product drastically. Photogrammetry applications deliver vast datasets with millions of points or polygons. For real time playback in VR environments these datasets cannot be used directly and need to be optimized. Aim of this research is to examine, compare and document different workflows in regards to factors relevant for production:

  • Photographic quality and precision of the output,
  • Real time performance of the model in VR applications,
  • Processing time and manual effort required,
  • Cost for software and hardware.
Triangulation is the basic principle behind the photogrammetric process where overlapping photos taken at different angles intersect. By mathematically modelling intersecting lines associated with features visible on the photographs, 3D points located in space are extracted. Through an bundle adjustment process, which includes processing the camera calibration and orientation as well as segmentation of the photographs, 3D point coordinates are extracted. To generate the final 3D objects, the 3D-point cloud goes through a meshing and texturing process.

Left: Raw mesh with 125.358 points reconstructed using Agisoft Photoscan. Right: Optimized intermediate mesh with 6.225 points auto-retopologized with 3D Coat.

The drawback of this technology is the high polygon count of the captured 3D objects. A big challenge during the game development process is the optimization of multiple assets which constitute the game and it is why only a few games are currently using it. Since the digital approach to photogrammetry is relatively new (the origin of analog photogrammetry reaches far back in history though) the production processes are barely documented. What is missing so far are direct and practical comparisons between different applications and workflows for specific subjects like indoor, outdoor or isolated objects especially on a practical level for Indie user with limited budget.

Automatically optimized and textured model with 1.729 points using Simplygon and the same model textured.

Examples of Retopologized Scans