Thanks to recent developments, gaze analysis and eye-tracking applications are now highly accurate and effective in many practical uses. This survey paper presents a comprehensive summary of work from the last five years, serving as a valuable single source for understanding the current state of the art. Bisogni et al. have taken the approach of reviewing all possible application areas, rather than surveying only gaze tracking in human-computer interaction (HCI), for example. This works well since the degree of technical commonality is striking.

The authors begin with a brief discussion and review of the fundamental challenges and techniques of eye tracking. The clear presentation of the classes of eye movements is important for technology-driven readers to understand the rest of the paper. They list many available eye-tracking devices, with their key attributes, in a key table; it’s noteworthy that (with one exception) the sample rates range from 30 to 200 Hz.

The authors consider recent gaze analysis as primarily occurring in three categories: medical, HCI, and virtual and augmented reality (VR/AR) applications. Studies have explored the use of gaze analysis in medical applications such as disease detection, clinical practice analysis, and the use of gaze information to aid in training medical image analysis networks. The survey in this area covers the field well, and conveys how much and how wide-ranging research is ongoing.

HCI is a commonly thought-of area for the application of gaze analysis, and the work surveyed bears this out. Much research looks at the link between gaze and attention, as well as learning outcomes. Related applications are being explored, not in learning but in marketing and commerce: unsurprisingly, there is great interest in understanding how screen activity and layout increases the likelihood of online purchase. In VR/AR settings, eye tracking allows for immediate and nonintrusive interaction. Interestingly, “in recent years, research has focused more on AR than VR regarding gaze analysis.”

For each of these three categories, current research is presented in very useful charts of the major areas, applications, and gaze sensors; the key datasets available; and timelines of recent developments. There are many promising new uses of gaze analysis beyond these three, for example, this survey calls out learning comprehension and driver behavior. There is also significant work on understanding and enhancing human performance--enough that it would be an appropriate fourth category. For example, recent work investigates the use of gaze analysis in the areas of athletic performance and in pilot readiness and performance training.

One sobering insight from the survey is the (still) limited amount of data available for algorithm development and verification. Many researchers are using techniques such as deep learning, which demand large amounts of data, yet most studies are constrained to data from 25 subjects or less. Augmentation and other usual techniques for the effective use of small datasets do not answer questions of inter-subject variations, age-related differences in gaze behavior, and other important issues. The authors note this as an open challenge: “If we consider how fine the pattern of the gaze is, especially under behavioral influences, [dataset size] numbers are not sufficient to provide a strong and universal guarantee of the methods.” The survey does not examine current patents in the area, perhaps by design, but a look at recent patents and patent applications would add to the view of the current state of the art. Nonetheless, this survey paper is an essential read for anyone working in or interested in gaze analysis and eye detection, whether as a researcher or a commercial implementer.