Improved Three-Dimensional Gaze Estimation with Precise Iris Segmentation Based on YOLOv8

This paper focuses on one of the highly discussed issues in the field of computer science and engineering—three-dimensional gaze estimation, which holds significant application value in virtual reality, augmented reality, medical applications, and other domains. Traditional gaze estimation methods, particularly those based on iris features, suffer from accuracy limitations in iris parameter acquisition. To overcome this limitation, this paper proposes a precise iris segmentation method based on YOLOv8, aiming to achieve high-precision three-dimensional gaze estimation under simplified device conditions. By constructing a diverse dataset and utilizing an improved YOLOv8 model, optimizations are made in the backbone network, feature pyramid structure, and other aspects to enhance the accuracy of iris image segmentation and, consequently, improve gaze estimation accuracy.