Modeling the gaze point distribution to assist eye-based target selection in head-mounted displays

Due to its natural and fast characteristics, eye tracking, as a novel input modality, has been widely used in head-mounted displays for interaction. However, because of the inadvertent jitter of eyes and limitations of eye tracking devices, the eye-based selection often performs poorly in accuracy and stability compared with other input modalities, especially for small targets. To address this issue, we built a likelihood model by modeling the gaze point distribution and then combined it with Bayesian rules to infer the intended target from the perspective of probability as an alternative to the traditional selection criteria based on boundary judgment. Our investigation shows that using our model improves the selection performance significantly over the conventional ray-casting selection method and using the existing optimal likelihood model, especially in the selection of small targets.

Automated summary

Eye tracking, as a novel input modality, is widely used in head-mounted displays for interaction due to its natural and fast characteristics. However, eye-based selection often performs poorly in accuracy and stability compared with other input modalities, especially for small targets. To address this issue, we built a likelihood model by modeling the gaze point distribution and combined it with Bayesian rules for probabilistic inference of the intended target as an alternative to traditional selection criteria. Our investigation shows significant improvement in selection performance, especially for small targets, compared to conventional methods and existing optimal likelihood models.