Real-time, accurate, and robust pupil detection is an essential prerequisiteto enable pervasive eye-tracking and its applications -- e.g., gaze-based humancomputer interaction, health monitoring, foveated rendering, and advanceddriver assistance. However, automated pupil detection has proved to be anintricate task in real-world scenarios due to a large mixture of challengessuch as quickly changing illumination and occlusions. In this paper, weintroduce the Pupil Reconstructor PuRe, a method for pupil detection inpervasive scenarios based on a novel edge segment selection and conditionalsegment combination schemes; the method also includes a confidence measure forthe detected pupil. The proposed method was evaluated on over 316,000 imagesacquired with four distinct head-mounted eye tracking devices. Results show apupil detection rate improvement of over 10 percentage points w.r.t.state-of-the-art algorithms in the two most challenging data sets (6.46 for alldata sets), further pushing the envelope for pupil detection. Moreover, weadvance the evaluation protocol of pupil detection algorithms by alsoconsidering eye images in which pupils are not present. In this aspect, PuReimproved precision and specificity w.r.t. state-of-the-art algorithms by 25.05and 10.94 percentage points, respectively, demonstrating the meaningfulness ofPuRe's confidence measure. PuRe operates in real-time for modern eye trackers(at 120 fps).
展开▼
