查看更多>>摘要：The following quote was obtained by the news editors from the background informa tion supplied bythe inventors: “Eye-tracking systems capture images of the eyes in order to determine the 3D gaze of theuser, or a 2D projection of that gaze onto a surface or plane, such as a screen or typical viewing distance.This is d one either through a computer vision segmentation of the image of the eye into v arious parts,i.e. pupil, sclera, iris, eye lids, canthus, etc., the features of which are then exported as parameters thatcan be used to calculate the user’s gaze based on calibration data or generate an eye model for the samepurpose, or the eye images are fed directly into a neural network or other machine learning approach thatinfers the segmentation and/or user’s gaze directly from the imag es based on a database of labeled eyeimages. The parameters extracted from a tr aditional computer vision approach can also be used with amachine learning appr oach, with or without the images of the eyes, which may also be scaled to variou slower resolutions. In all cases, the quality of the images, with respect to co ntrast, lighting, sensitivity, etc.,and the amount of computation required to e xtract the features of the eye or infer the gaze directly fromthe images is of first importance to the robustness and quality of the gaze estimate. This is esp eciallytrue in a head-mounted, mobile system intended to operate both indoors a nd outdoors, in uncontrolledand variable lighting conditions. The complexity of extracting information from the eye images, especiallythe crucial pupil positi on, requires high complexity in the computer vision algorithms used for the task ,and robustness to environmental effects on those images is the main challenge remaining for eye-trackingsystems.”
NETL
NSTL