Computer Vision for Thinking Heads Face and Facial Feature Tracking

1
Computer Vision for Thinking Heads - Face and
Facial Feature Tracking

• Roland Göcke, NICTA ANU, Canberra
• Trent Lewis David Powers, Flinders University,
  Adelaide

2
Computer Vision

• Computer vision is the science and technology of
  machines that see. (Wikipedia)
• Computer vision seeks to apply the theories and
  models of computer vision to the construction of
  computer vision systems (Wikipedia)
• Hardware (camera systems) and software
  (processing of images and videos)

3
Computer Vision (cont.)
4
Applications

• Face detection
• Face and facial feature tracking
• Gesture / expression recognition (face, body,
  hand)
• Visual surveillance
• Object (e.g. people) tracking

5
Face Detection

• Statistical model of skin colour (Jones Rehg,
  1999)
• Choosing the right colour model facilitates
  independence of (perceived) skin colour (HSV,
  CIELAB)
• Another model uses Haar-like features (simplified
  Haar wavelets) to learn the composition of faces
  from these features (Viola Jones, 2001)
• Fast
• Freely available implementation in Intel OpenCV
  library

6
Facial Feature Tracking

• Active Appearance Models (AAM) combine
  statistical shape and texture models (Cootes,
  Edwards Taylor, 1998)
• A parametric model of non-rigid visual objects

7
AAM (cont.)

• Alignment based on finding model parameters that
  fit learned model to image

8
Stereo Vision

• Two cameras viewing scene similar to biological
  vision
• If system is calibrated, the 3D position of
  object points (e.g. on the face) can be recovered
• Improved accuracy and robustness for head poses
  not frontal to cameras

9
Vision Beyond the Visible Spectrum

• Far infrared (thermal) 7,000 15,000nm

• Near infrared 900 1700nm
• Specific wavelength filters for haemoglobin and
  water

10
Conclusions

• Many useful computer vision algorithms exist for
  face and facial feature tracking
• Three core issues
• Accuracy
• Robustness (pose, occlusion, illumination)
• Speed
• Future
• New camera technology that sees beyond the
  visible spectrum