Compensating for camera translation in video eye movement recordings by tracking a landmark selected automatically by a genetic algorithm

We develop a method for accurately estimating the motion of a camera relative to a highly deformable surface, specifically the movement of a camera relative to the eye. A small rectangular landmark is selected and tracked throughout a set of video frames as a measure of vertical camera translation. The specific goal is to present a process based on a genetic algorithm that selects a suitable landmark. We find that co-correlation, a statistic relating the time series of a large population of landmarks, is a robust predictor of the accuracy of the landmarks. This statistic is used to iteratlvely select the best landmark from the population. At each iteration new landmarks are created that inherit properties of the previous population of landmarks. We show that the algorithm can select a landmark that will estimate camera translation with an accuracy of 1.8 pixels, which means that the direction the eye is looking can be determined with an accuracy of better than 0.6°.