Contact address logo Eisko, 24 rue de l'Est 75020 Paris
Contact phone logo +33 (0)1 85 08 52 51
Contact mail logo
> contact us!
Eisko logo


Tech Friday #1: Yet another capture system? Or: why photogrammetry just isn't good enough!

November 21, 2014

Hi Folks,

In our first Tech Friday post, we wanted to give you a brief insight into the motivations and approach underlying our capture system.

Since their introduction for computer assisted design and quality inspection, a number of different approaches to capturing 3D objects have been investigated:

    scanner scan                                      contact measurment                           structured light

These approaches can be classified according to:
  • The physical properties they can acquire:  '3D scanning' generally refers to geometry capture, potentially supplemented by photos mapping.
  • Wheter capture is passive or active: does the capture process modify/interfer the phenomena it records ?
  • The type of object captured: does the process record static or mobile/articulated/deformable objects ?
While contact measurments and manual scanner are not well adapted to capture non-rigid and deformable objects like the face, structured-light and photogrammetry are potential good candidates for geometry capture.

Photogrammetry is based on the detection and registration of similar features in multiple DSLR images. External and internal calibration of the camera sensors enables to triangulate the associated position in space.

Such capture rigs are easy to set-up and progressively more affordable, thanks to the increasing availability of good-quality cameras. This approach enables to deal with a subject in different poses, since shooting is almost instantaneous.

But looking closer to it, the resulting scan is noisy and unstructured...

It took us some time to realize that such artefacts result from two fundamental causes, independent of the number and resolution of the cameras used:
  • Light scatering caused by specular and subsurface properties of the skin, surface fuzz, eyelashes which reflects and absorb lights depending on the direction so that epipolar features can not be identical in different camera views.
  • The difficulty to achieve a perfectly accurate geometric calibration of the cameras - which still remains an open problem for photogrammetry.
After years of technology survey and experimentations on this thematic, we finally decided to develop our own capture system. Our objective was to ensure the simultaneous acquisition of both detailed geometry and physically-based textures which characterize the precise appearance of any subject's pose.

To do so, we do not consider a direct acquisition, but analyze and dissociate the way in which the skin, the mouth, the eyes diffuse, reflect and absorb light under precisely calibrated illumination conditions.

Our capture system is made of 1.600 LED lights which permit us to control the intensity and direction of the lighting very accurately. Our high-frequency 'material cameras' enable us to separate the diffuse and specular components of the surfaces being recorded, and extract normal information for each RGB color channel, among other things.

The result of this analysis is a clean 3D geometric scan, resolution-independent down to 20 micrometers, complete with tiny details like wrinkles and pores, totally free from reflection and refracton side effects traditionnal in photogrammetry :)

I hope this post has helped you understand why we decided to build another capture system. In future Tech Fridays, we'll look at what makes Eisko unique both on the capture and reconstruction sides.