Fast and robust pupil detection is an essential prerequisite for video-based\neye-tracking in real-world settings. Several algorithms for image-based pupil\ndetection have been proposed, their applicability is mostly limited to\nlaboratory conditions. In realworld scenarios, automated pupil detection has to\nface various challenges, such as illumination changes, reflections (on\nglasses), make-up, non-centered eye recording, and physiological eye\ncharacteristics. We propose ElSe, a novel algorithm based on ellipse evaluation\nof a filtered edge image. We aim at a robust, resource-saving approach that can\nbe integrated in embedded architectures e.g. driving. The proposed algorithm\nwas evaluated against four state-of-the-art methods on over 93,000 hand-labeled\nimages from which 55,000 are new images contributed by this work. On average,\nthe proposed method achieved a 14.53% improvement on the detection rate\nrelative to the best state-of-the-art performer.\ndownload:ftp://emmapupildata@messor.informatik.unituebingen. de\n(password:eyedata).\n