Visual Face Tracking and Recognition

Visual object tracking is often used in robotics for various applications such as visual navigation, human following, object grasping and manipulating, gesture recognition, visual surveillance. A robot must store a large amount of visual information to be able to operate in the environment, recognize and track objects and persons. The purpose of this project is to develop an effective way to use a low-redundancy image representation for robust automatic tracking and recognition of objects or structures appearing in images.

Research focuses on the following topics:
  • Fast image decomposition: A low-redundancy representation must be able to appropriately identify and store the significant object features which is important for robust object and face recognition and tracking like the well known caricature effect where we are able to recognize a face from few specific but distorted features. The object image can be reconstructed from the wavelet representation.
  • Face tracking: The resulting Gabor representation of an object image can be used as object model. Positions of particular wavelets can be changed by varying the translation, rotation, scale and other model parameters.
  • Face recognition
The original (left) and the reconstructed images with the following number of Gabor wavelets (from left to right): 28, 85, 150, 250, 450, 650.
Wavelet based tracking a face image.

Contact

Dr. Alexander Mojaev, Tel.: (07071) 29-78982, alexander.mojaev (at) uni-tuebingen.de