Gaze-contingent visual display systems connect an eye- or gaze-tracking system with a display of the visual stimulus, to change the displayed image as a function of where the viewer looks. Since their introduction almost 40 years ago, they have been used to address many areas of vision research. They have also been used in clinical assessment devices and other medical applications, including rehabilitation. Ideally, the observer's view would be instantaneously updated in response to gaze changes. However, all such systems introduce a lag between the gaze change and the view update: the system latency. If the system latency is sufficient, it may interfere with the gaze-contingent experience The most difficult situation for most gaze-contingent display systems occurs with saccadic (very fast) eye movements. With current best practices, delays within the components of the system can lead to substantial mismatches between gaze and stimulus, which are worse with larger eye movements. Recently, we developed a compressed-exponential model to predict saccadic eye movements, allowing the display to be updated based on the prediction. Saccade prediction, when combined with knowledge of system latency, produces smaller errors in gaze-contingent displays. As we developed that model, we found saccades that had a curved path (assumption is straight) or a non-conventional velocity profile (expected pattern is approximately symmetric rise then fall). Such saccades cause additional display errors and are not expected given the ballistic nature of saccades.
Aim 1 : To investigate conditions under which these non-standard saccades occur due to curved paths and non-conventional velocity profiles. This will inform understanding of saccade generation and control.
Aim 2 : To develop an algorithm that is robust to departures from prior predictions by combining the Kalman filter with our proven compressed-exponential model. By correcting for curved saccade paths, non-conventional accelerations and other deviations from the expected trajectory, these prediction algorithms will reduce or eliminate update lag. The effect that update lag has on visual task performance is not well known, and therefore the extent of the problem in past gaze-contingent experiments has not been determined.
Aim 3 : To evaluate the effects of saccadic eye-movement prediction to compensate for system latency in three studies that use gaze- contingent displays in different ways. The studies will be based on popular gaze-contingent paradigms, and will reveal how the lag affects each of them. The methods developed here are expected to become the standard approaches for controlling displays that respond to saccadic eye movements, thereby having wide impact in many fields, from reading to robotics to rehabilitation. This project is a collaboration between Russell Woods and Gang Luo, at the Schepens Eye Research Institute, Massachusetts Eye and Ear, Harvard Medical School.
Gaze-contingent systems that change the image on the display according to where the viewer looks are used in many clinical, research and applied settings, including retinal assessment, reading, rehabilitation and robotics. Some systems exhibit significant lag between the eye movement and the display update, and there is evidence that the experience is affected. We will develop a method to compensate for this lag, and demonstrate the effects of system lag on experiments with gaze-contingent systems.
|Wang, Shuhang; Woods, Russell L; Costela, Francisco M et al. (2017) Dynamic gaze-position prediction of saccadic eye movements using a Taylor series. J Vis 17:3|