Appearance-based gaze tracking with spectral clustering and semi-supervised Gaussian process regression

Ke Liang (1) , Youssef Chahir (2) , Michele Molina (3) , Charles Tijus (4) , François Jouen (4)

1 Laboratoire de Psychobiologie du Développement
2 Equipe Image - Laboratoire GREYC - UMR6072
3 LPCN - Laboratoire de psychologie de Caen Normandie
4 CHART - Cognitions Humaine et ARTificielle

Two of the challenges in appearance-based gaze tracking are: 1) prediction accuracy, 2) the efficiency of calibration process, which can be considered as the collection and analysis phase of labelled and unlabelled eye data. In this paper, we introduce an appearance-based gaze tracking model with a rapid calibration. First we propose to concatenate local eye appearance Center-Symmetric Local Binary Pattern(CS-LBP) descriptor for each subregion of eye image to form an eye appearance feature vector. The spectral clustering is then introduced to get the supervision information of eye manifolds on-line. Finally, taking advantage of eye manifold structure, a sparse semi-supervised Gaussian Process Regression(GPR) method is applied to estimate the subject's gaze coordinates. Experimental results demonstrate that our system with an efficient and accurate 5-points calibration not only can reduce the run-time cost but also can lead to a better accuracy result of 0.9°.

Mots clés: spectral clustering; Appearance-based gaze estimation; Gaussian process regression

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]; Traitement des images [eess.IV]; Intelligence artificielle [cs.AI]; Apprentissage [cs.LG]; Multimédia [cs.MM]

Ke Liang, Youssef Chahir, Michele Molina, Charles Tijus, François Jouen. Appearance-based gaze tracking with spectral clustering and semi-supervised Gaussian process regression. Conference on Eye Tracking South Africa (ETSA 2013), Aug 2013, Le Cap, South Africa. pp.29-31, ⟨10.1145/2509315.2509318⟩. ⟨hal-01882823⟩