Gaze Estimation

Gaze estimation studies how to infer a person's viewing direction or target from images, video, and eye-tracking sensors, supporting applications such as driving, robotics, virtual reality, and human-computer interaction. Methods range from geometric eye models and calibration to deep appearance-based models that combine eye and head features with scene context; newer approaches also use temporal prediction, vision-language models, and synthetic or unlabeled data. A central challenge is reliability across people, devices, lighting, movement, and unfamiliar settings, where calibration needs, label noise, and demographic disparities can undermine accuracy. Current work also weighs predictive performance against real-time compute, uncertainty reporting, privacy, and the need to validate systems in practical environments.