Today it's possible to create advanced eye tracking systems without specialized hardware but simply using a standard webcam and the inevitable AI. But the real difference is made by the model that is already completely on the edge and therefore on the device or browser, allowing you to work even offline!
With 'AI on the edge' technologies for facial and eye detection, a bit of linear regression, and a real-time calibration system that adapts to the user and environment, I will show, using only JavaScript and on-device AI models, the results of the experiment, discussing accuracy, current limitations, and potential improvements.
The challenges faced in on-device optimization are many to balance precision and performance, but this experiment opens new possibilities for accessible applications in human-machine interaction, behavioral studies, and accessibility systems, all created with open source technologies and common hardware!
This talk has been presented at JSNation US 2025, check out the latest edition of this JavaScript Conference.