Probabilistic Eye Contact Detection for the Robot-assisted ASD Diagnostic Protocol

Abstract

This paper describes a probabilistic method for eye contact detection, aimed at facilitating the diagnostic process of Autism Spectrum Disorder (ASD). Eye contact, which is a special case of a wider social-attention cue called gaze, plays a major role in ASD diagnostics protocols used in clinical practice. Therefore, a reliable method for automatic eye contact detection is a key capability to open the way towards robotassisted autism diagnostics, which has the potential to reduce diagnostic time and increase reliability. The proposed method uses data from a simple low-resolution monocular camera, which is built into the NAO humanoid robot, and uses head pose and gaze direction as its inputs, which are very prone to outliers. We use a probabilistic framework in order to provide continuous measure of the eye contact and increase robustness to outliers. Furthermore, we take into account the temporal aspect of eye contact, in order to discard short glances which do not indicate actual transfer of attention. Initial experimental results, conducted in a laboratory setting, confirm that the proposed method can be effective in detecting eye contact with the NAO humanoid robot.

Files

Full Paper as PDF

BibTeX Citation

DOI

10.20532/ccvw.2016.0001

https://doi.org/10.20532/ccvw.2016.0001

BibTeX

@InProceedings{10.20532/ccvw.2016.0001,
  author =       {Petric, Frano and Mikli{\' c}, Damjan and Kova{\v
                  c}i{\' c}, Zdenko},
  title =        {Probabilistic Eye Contact Detection for the
                  Robot-assisted ASD Diagnostic Protocol},
  booktitle =    {Proceedings of the Croatian Compter Vision Workshop,
                  Year 4},
  pages =        {3-8},
  year =         2016,
  editor =       {Lon{\v c}ari{\' c}, Sven and Cupec, Robert},
  address =      {Osijek},
  month =        {October},
  organization = {Center of Excellence for Computer Vision},
  publisher =    {University of Zagreb},
  abstract =     {This paper describes a probabilistic method for eye
                  contact detection, aimed at facilitating the
                  diagnostic process of Autism Spectrum Disorder
                  (ASD). Eye contact, which is a special case of a
                  wider social-attention cue called gaze, plays a
                  major role in ASD diagnostics protocols used in
                  clinical practice. Therefore, a reliable method for
                  automatic eye contact detection is a key capability
                  to open the way towards robot-assisted autism
                  diagnostics, which has the potential to reduce
                  diagnostic time and increase reliability. The
                  proposed method uses data from a simple
                  low-resolution monocular camera, which is built into
                  the NAO humanoid robot, and uses head pose and gaze
                  direction as its inputs, which are very prone to
                  outliers. We use a probabilistic framework in order
                  to provide continuous measure of the eye contact and
                  increase robustness to outliers. Furthermore, we
                  take into account the temporal aspect of eye
                  contact, in order to discard short glances which do
                  not indicate actual transfer of attention. Initial
                  experimental results, conducted in a laboratory
                  setting, confirm that the proposed method can be
                  effective in detecting eye contact with the NAO
                  humanoid robot.},
  doi =          {10.20532/ccvw.2016.0001},
  url =          {https://doi.org/10.20532/ccvw.2016.0001}
}