Eye detection in video can be performed quickly and robustly using ASEF filters.  Two filters are trained on thousands of images to produce an optimal output for each  eye.  Localization involves convolving the face detection region with the trained filters and selecting the maximum response.  Filter based eye detection is very fast and can locate eyes in under 4ms on modern processors.  Source code for this eye detector can be downloaded from the PyVision project or as a stand alone application called FaceL.

For more information see the paper:  D.S. Bolme, B.A. Draper, and  J.R. Beveridge. Average of Synthetic Exact Filters. Computer Vision and Pattern Recognition. June 2009.  (PDF) (Licensing)