Automatic detection of camera translation in eye video recordings using multiple methods

In video eye tracking, shifting of the camera relative to the head can introduce artifacts. This work proposes a new combination of image processing techniques to automatically detect and measure the relative translation of cameras separately imaging the left and right eyes. It uses a priori physiological knowledge to improve the accuracy of algorithms. The first method compares an eye image with a reference frame using cross-correlation methods. The second isolates the upper eyelid and compares it with a reference frame to improve approximation of camera translation. The third creates an eyelid template from multiple frames and cross-correlates it with each image frame. The later has the highest accuracy, with a mean error of 1.3 pixels. It is more robust since it eliminates features of the image that may introduce errors. This excellent accuracy makes the method a viable solution for the problem of camera movement relative to the head.