Estimating Point of Regard with a Consumer Camera at a Distance

Abstract

Abstract. In this work, we have studied the viability of a novel technique to estimate the POR that only requires video feed from a consumer camera. The system can work under uncontrolled light conditions and does not require any complex hardware setup. To that end we propose a system that uses PCA feature extraction from the eyes region followed by non-linear regression. We evaluated three state of the art non-linear regression algorithms. In the study, we also compared the performance using a high quality webcam versus a Kinect sensor. We found, that despite the relatively low quality of the Kinect images it achieves similar performance compared to the high quality camera. These results show that the proposed approach could be extended to estimate POR in a completely non-intrusive way.

Unobtrusive and pervasive video-based eye-gaze tracking

“…(e) Point-of-regard calibration determines parameter values that position the gaze on a monitor screen [22,25,27,37,40,45,56,57,61,62,[70][71][72][73][78][79][80][81][82][83][84][85][86][87] or on a planar surface positioned at a distance from the user in physical space [43,68,69,[74][75][76][77], hence permitting the estimation of a point-of-regard (PoR). This may be carried out by:…”

“…(ii) Implicit mapping which determines a mapping function that estimates the PoR by mapping the image contents directly to screen coordinates. This may be achieved by means of machine learning techniques [22,37,38,40,43,45,56,57,61,62,[82][83][84][85][86][87], or by mapping the distance of a specific eye region feature from a reference image point onto a corresponding fixation point on a monitor screen [25,27,[78][79][80][81]. Figure 4(b) illustrates this calibration procedure.…”

“…(e), the user is either required to maintain a frontal eye and head pose [68], or perform specific eye [25,27,37,40,43,45,49,56,57,61,62,64,[69][70][71][72][73][79][80][81][82][83][84][85][86] or head movements [62,64,69,77] by fixating at markers appearing sequentially on a monitor screen [25,27,40,45,49,56,57,61,62,64,[70][71][72][73][79][80][81][82][83][84]…”

“…Most of the overviewed video-based methods focus mainly on the application of eye-gaze tracking as a desktop technology. Hence, most research effort in past years has been mainly dedicated towards the development of methods that fit controlled conditions where the head movement, if any, is mostly constrained to a small volume [22,35,45,49,57,75,78,79,[81][82][83][84]89], the illumination conditions are stable or controllable by the projection of visible [62] or IR [29,31,57,90] illumination, and both calibration [22, 25, 27, 35, 43, 47, 49, 64, 64, 69-80, 88, 91] or the collection of a training data set [37,38,40,43,45,52,56,57,61,62,[82][83][84][85][86]92] may be carried out as required. Such conditions limit the applicability of these video-based eye-gaze tracking methods within the uncontrolled settings associated with pervasive eye-gaze tracking and, hence, call for methods that address the challenges which emerge from these limitations.…”

“…In comparison with the image quality captured by higher-end cameras used in commercially available eye-gaze trackers, consumer-grade cameras typically acquire images of inferior quality and wider field-of-view, which capture an expansive portion of the background. If the camera is placed at a distance from the user [37,38,40,81,82,84,86,[93][94][95][96][97][98][99] rather than mounted close to the eye regions [85,100,101], its wide field-of-view accommodates larger head movement that is advantageous within the context of pervasive eye-gaze tracking, in comparison with the narrower field-of-view that is typically associated with commercial eye-gaze trackers. Nonetheless, this reduces the perceived resolution of the eye region images, which calls for methods that address the challenge of estimating the eye-gaze from low-quality images.…”