The Cone Model: Recognizing gaze uncertainty in virtual environments

Abstract

Eye tracking is used by psychologists, neurologists, vision researchers, and many others to understand the nuances of the human visual system, and to provide insight into a person’s allocation of attention across the visual environment. When tracking the gaze behavior of an observer immersed in a virtual environment displayed on a head-mounted display, estimated gaze direction is encoded as a three-dimensional vector extending from the estimated location of the eyes into the 3D virtual environment. Additional computation is required to detect the target object at which gaze was directed. These methods must be robust to calibration error or eye tracker noise, which may cause the gaze vector to miss the target object and hit an incorrect object at a different distance. Thus, the straightforward solution involving a single vector-to-object collision could be inaccurate in indicating object gaze. More involved metrics that rely upon an estimation of the angular distance from the ray to the center of the object must account for an object’s angular size based on distance, or irregularly shaped edges - information that is not made readily available by popular game engines (e.g. Unity© /Unreal© ) or rendering pipelines (OpenGL). The approach presented here avoids this limitation by projecting many rays distributed across an angular space that is centered upon the estimated gaze direction.