Eye Tracking For Single-User Stereo
September 22, 2011, 3-D User Experience Technical Summit, Hollywood–Ihor Petelycky from SpecialView described eye tracking technologies to improve single-user 3-D autostereoscopic displays. He reviewed the operation and characteristics of these display types and looked at two implementations.
Displays and software for 3-D viewing are improving at a high rate. To meet the new requirements, manufacturers are working on auto-stereo displays. The main types of displays are lenticular, parallax barrier, and volumetric which are 2-view for a single user and multi-view for multiple viewing zones. Implementations for tunable lenticular films over a standard display allow the stereo image to be available in a range of viewing positions and angles depending on the viewing distance.
By introducing eye tracking, the computer can enhance the image to optimize the viewing area. This frees the user from having to find the sweet spot for the images and hold that position. This task that is not easy or comfortable for viewing times over 20 minutes. The difference between small screens for hand-held devices and larger screens for tethered operations are significant. The challenges of crosstalk and ghosting are harder to correct for the larger screens.
Overall, content matters and the adaptation for any screen requires that you start with good 2-D. The color temperature, resolution, and other image properties need to be of the highest quality before processing to 3-D. The conversion process is non-linear and adds its own aberrations to the images on a conversion per device basis.
One implementation is to use a lenticular overlay on a laptop screen. A detailed alignment and calibration process starts with the user setting interocular distance and some other software optimizations. Software adjusts for coarse air gap, tilt, shift, and dot pitch. This processes is repeated until the image is considered acceptable and needs to achieve sub-pixel accuracy. Then the eye tracking process needs to be trained for the particular camera and available CPU processing power.
Unfortunately, most webcams are not designed for 3-D tracking. They have lag time and low light errors. The eye tracking, once trained, can accommodate user movement and will convert the images back to 2-D if the user is out of range. Overall, the system needs total integration to identify the face and position of the eyes. It requires personalization and calibration to handle the gaze tracking and high resolution location. In addition, it needs to fail gracefully to ease visual strain.
Limitations in the systems include high error factors in lighting and false face recognition. The various components need to be integrated with the 3-D rendering engine and puts a load on the compute resources in the computer. Future systems are in development to incorporate IR tracking to increase accuracy. Ways to increase system integration include multiple cameras and better feature recognition.
