Facial Scanning

Acquisition of geometry and texture data through 3-D scanning systems can provide access to very realistic models of real persons. 3-D facial scanning is popular for visual effects productions and computer games based on real-life talent (for example sports games). This article describes production experiences with 3-D scanners and provides tips and tricks on post-processing the acquired meshes.

Table of contents

Scanning Equipment

Hardware

  • InSpeck 3-D (http://www.inspeck.com/). We have used a version of this scanner (courtesy of Sony Computer Entertainment America (http://www.scea.com/)) to acquire data for several heads with good results. The model data required extensive cleanup to achieve good topology. The texture resolution on the model we tested was about 1.3 megapixels and required additional reference photography for good results.
Facial scanning setup with InSpeck 3-D scanner.
Enlarge
Facial scanning setup with InSpeck 3-D scanner.
Facial scanning setup with InSpeck 3-D scanner.
Enlarge
Facial scanning setup with InSpeck 3-D scanner.


Software

  • Geometry System Studio (GSI Studio) (http://www.geometrysystems.com/). We have worked on heads acquired with InSpeck's 3-d scanner and processed with GSI Studio version 3.0 beta 4.0.
  • Re-meshing and additional data cleanup can be performed directly in Softimage (http://www.autodesk.com/softimage).



Data Acquisition


3-D Scanning of Base Pose

Base Pose: relaxed jaw muscles (mouth slightly open), eyes open looking straight ahead.
Enlarge
Base Pose: relaxed jaw muscles (mouth slightly open), eyes open looking straight ahead.

The base pose acquired should allow for easy post-processing (remeshing, model cleanup, insertion of interior of mouth and eye sockets).

Using the InSpeck 3-D scanner, we took scans from eight poses acquired at 45-degree intervals plus one each for the top of the head and the underside of the neck. They were assembled into a single, high-resolution mesh.

Mouth open or closed? We have scanned with mouth open, in a pose that corresponded to a completely relaxed jaw muscle.

Eyes open or closed? We have scanned with eyes open.



3-D Scanning of Orthogonal Poses

To generate reference material for the subsequent soft tissue tuning, we also acquired six additional "orthogonal facial poses" that exercise the most relevant facial deformations. These poses were acquired with eyes open. The orthogonal poses were recorded with two simultaneous scans that were recorded at approximately 80 degrees to each other, or about plus/minus 40 degrees to the left and right from the center plane of the face. By fusing these together we got good data of the front of the face in all poses.


(1) Smile: raw scan data of a wide smile pose
Enlarge
(1) Smile: raw scan data of a wide smile pose
(2) Scream: raw scan data of a scream pose
Enlarge
(2) Scream: raw scan data of a scream pose
(3) Squint: raw scan data of a squinting pose
Enlarge
(3) Squint: raw scan data of a squinting pose
(4) Puff: raw scan data of a wide cheek puffing pose
Enlarge
(4) Puff: raw scan data of a wide cheek puffing pose
(5) Tendons: raw scan data of a tendon straining pose
Enlarge
(5) Tendons: raw scan data of a tendon straining pose
(6) Pursed: raw scan data of a lip pursing pose
Enlarge
(6) Pursed: raw scan data of a lip pursing pose


Model Cleanup

  • Re-meshing: In order to convert the scanned head to a usable model for Facerobot it must be converted to a much lower poly mesh with good edge flow to support deformation.
  • Identifying or Creating a good Edgeflow template: The first step is identifying a edgeflow template that works well with your head, for example a man with 3 chins will require more edgloops and detail in the chin area than a skinny person so a different edgeflow template should be used.
  • Template Placement: Once you've decided on a good template, the next phase is to positon the template in the same physical space as the scanned head, using the proportional or brush tool move the template areas so it conforms to the scanned data as closely as possible, making careful note to maintain good parameterization and have the edgeflow follow the models anatomical landmarks.
  • Shrink-Wrapping Template: Once this is done shrinkwrap the edgeflow template to the scanned data using the closest vertex option. Free the template and perform and overall relax to smooth out the topology. there will be problem areas with the shrink wrap usually around the eye areas(due to eyelashes) and nose/mouth. You will need to go in manually and adjust the vertices around these points, in most cases remodelling of these areas will be required. Keep iterating using relax and shrink wrap and modelling locally on areas of the model until you have the best fidelity reproduction of the scanned data. Areas such as the ears usually will not work with this shrink wrap process due to the many folds so it may be necessary to manually place these on your model then perform a subtle shrink wrap to have it conform better to the scanned data ears.
  • After re-meshing, the model still needs to be completed (eye sockets, eye region detail, interior of mouth) to be used with Face Robot.

This page was last modified 18:20, 6 Aug 2009.
This page has been accessed 9701 times.

© Copyright 2009 Autodesk Inc. All Rights Reserved. Privacy Policy | Legal Notices and Trademarks | Report Piracy