EllSeg: An Ellipse Segmentation Framework for Robust Gaze Tracking

Real-Time Gaze Tracking with Event-Driven Eye Segmentation

“…Feature Extraction While historically hand-crafted, geometric features are popular [2,34,41,77], DNN algorithms have recently been shown to be more robust and accurate [20,33,53,56,59,86]. In particular, extracting features through eye segmentation [20,53,56,86] is by far the most widely used method.…”

“…Our algorithm uses a similar approach as DeepVOG [86], RITnet [20], OpenEDS [6], and Kim et al [53], known as "part segmentation," which segments the eye image into four parts: pupil, iris, sclera, and background. In contrast, Ellseg [56] and Wang et al [81] use elliptical segmentation, which predicts the ellipses of the pupil and iris.…”

“…Gaze tracking is critical to many fields [65], such as medicine [13,15], human-machine interaction [19,49,66,76], psychology [67], augmented reality (AR) [48,72], and virtual reality (VR) [23,46,71,83,88]. Due to its importance, there has been much research improving the tracking accuracy, both using conventional infrared/RGB cameras [20,50,53,56,86] and emerging event cameras [12].…”

“…Efficiency-Oriented Segmentation Coupled with the Auto ROI mode, we propose a lightweight neural network-based algorithm to predict the gaze from an ROI. Like much of the recent work [20,53,56,86], we use a model-based approach, where we first parameterize an eye image through eye segmentation, which is then used to estimate the gaze through fitting a geometric eye model.…”

The Development of Long-Distance Viewing Direction Analysis and Recognition of Observed Objects Using Head Image and Deep Learning

“…The advantage is that they can accurately judge the sight direction within the effective distance, and the head rotation does not affect it. In terms of image analysis, since the ellipse shape detects the position of the pupil or iris, a convolutional neural network can be used to directly segment the entire ellipse structure to deal with image occlusion caused by cameras and eyelashes, for example [16]. However, because the effective content is only about 50 to 80 cm, and the cost of such instruments is too high and challenging to get, it is not easy to popularize.…”