In general, the eye tracking latency works out to be the sum of: Eye camera exposure time (~8 ms) + USB transmission time (~8 ms) + processing time (~4 ms). Therefore, the current total latency of FOVE Eye Tracking is about 20 ms.
In practice, the total latency becomes: ~20 ms + however long it takes the application to use the eye tracking data and update
This is more in the range of ~40-50 milliseconds for a complete turnaround from eye movement, to application reaction, to a new image being displayed on the screen. However, the pure eye tracking data is available much sooner than the complete turnaround time.
Various factors can affect the latency—particularly the OS and UVC camera driver. We have seen situations where the Windows UVC driver emits two frames at a time. The eye tracking timestamps emitted from the FOVE SDK (as of v1.0) are taken upon receiving the camera frame from the OS.
Our eye tracking algorithm uses some internal filtering, but this is kept as minimal as possible to minimize latency. As of v1.0, no predictive algorithms are in use by default.
A third party research paper by Niklas Stein also provides measurements of FOVE's latency, as well as comparisons with other HMD eye trackers.
https://journals.sagepub.com/doi/full/10.1177/2041669520983338
Japanese Version: