Eye Tracking and locating is very important in autostereo display. When the system get the precise location of the double eyes and adjust the according parameters, the viewer can have the true feeling of stereo scene. The article uses a general to specific way to track the face and then locate the eye precisely. The red eye effect can help choose the candidates of eye regions. We use Adaboost to detect face and eye. In the face tracking part, we use the Adaboost cascade result to generate the likelihood model in particle filtering. And last the projection method is used to find the location of the pupil. The experiment shows that this method has greatly improved the speed of eye tracking and locating to 100f/s and well satisfied the requirements of the system.