QUERY DRIVENLOCALIZEDLINEARDISCRIMINANTMODELS FOR HEAD POSE ESTIMATION

Junsong Yuan

Abstract

Headposeappearances underthepanandtilt variations spana highdimensional manifold thathascomplex structures andlocal variations. Forposeestimation purpose, weneedtodiscover thesubspace structure ofthemanifold andlearn discriminative subspaces/metrics forheadpose recognition. Theperformance oftheheadposeestimation is heavily dependent ontheaccuracy ofstructure learnt and thediscriminating powerofthemetric. Inthis workwe develop a querypoint driven, localized linear subspace learning method that approximates thenon-linearity ofthe headposemanifold structure withpiece-wise linear discriminating subspaces/metrics. Simulation results demonstrate theeffectiveness oftheproposed solution in bothaccuracy andcomputational efficiency.
BibTeX
@inproceedings{yuan2007query,
  author = {Junsong Yuan},
  title = {{QUERY DRIVENLOCALIZEDLINEARDISCRIMINANTMODELS FOR HEAD POSE ESTIMATION}},
  booktitle = {ICME},
  year = {2007},
  doi = {10.1109/icme.2007.4285024},
}

Loading related content...

Figure