My Experiences with an ASL Mobile Eye Tracker
Background
To test my Industrial Design thesis on Campus Emergency Notification Devices, I wanted to put a prototype in front of some real users. After running a few walk-through tests, I realized that I needed a way to record and analyze these tests. I contacted the Center for Human Computer Interaction at Virginia Tech to inquire about their eye tracking equipment. I was given access to an ASL Mobile Eye tracking device. The Mobile Eye allows the user to wear a recording device that allows them full range of movement instead of being tied down to a computer. This was perfect for my use since I needed users to be able to walk around a room and evaluate my prototype.
Software
The Mobile Eye hardware comes with proprietary software that allows for calibration and video capture. I found this software to be capable, but pretty outdated. As an interaction designer, I found a lot of flaws but was still able to get everything setup. The first step for calibration is to align the user's eye in the center of the eye camera. This is relatively simple to do but is very important in maintaining accuracy later during capture. The second step is to adjust the sensitivity settings in the software to get a good reading of the three infrared beams that reflect off the surface of the eye. These beams are used to located the user's pupil (yellow circle). The last step is to instruct the users to look at several different points in front of them and indicate to the software where they are looking. This gaze calibration is done with 6-9 points in order to get a good, accurate reading.
Once all of this calibration is complete, the software will show a real-time view of what the user is looking at. This can be recorded straight from the computer or used to playback footage that was captured walking around.
Example
I'm keeping my thesis confidential to pursue potential patent applications but I wanted to show some actual footage shot by myself when I was getting familiar with the equipment. I decided to record myself and a few friends playing Xbox because gaming offers a lot of constant eye movement and scanning. Xbox also offers a relatively narrow "cone of focus" that will stay within the bounds of the eye tracking glasses and keep the capture highly accurate.
After capturing about 10 minutes of gameplay, I started to see results that backed up a lot of assumptions I had about different styles of gameplay. In each of the videos, you can see that the player will do a modest amount of scanning, but mostly keeps their gun pointed in the direction they are looking. This is to minimize the reaction time of having to swing the gun over to a target that is identified. By moving the gun with the eye, this reaction time is greatly increased. In the second video, you can also see how much the mini-map is utilized when it's available.
Limitations
Throughout my testing I found a lot of limitations that I'd like to share with anyone that is considering using this model eye tracker. During the testing of my prototype, I had a lot of issues with calibration once the subject was untethered from the computer. I think that in the process of walking around the subject's eye shifted and the alignment in the eye camera suffered. This caused issues with accuracy where ever the subject looked.
The cone of focus I talked about earlier also caused problems that were specific to the way I setup my tests. The Mobile Eye is accurate when the subject is looking forward and doing light scanning. It can't track the pupil when the subject looks out of the corner of their eye. This happened frequently during my tests because the normal location for network clocks and other notification devices is located near the ceiling at the top of the wall. My subjects ended up scanning for my prototype out of the top of their eye which was either on the fringe of where the Mobile Eye could track or completely outside of its range.
These two issues made me rethink how my tests were setup and how to brief the user before each test. More re-iteration on just how sensitive the eye tracking equipment was needed to make sure the subject didn't scratch their eye or adjust the glasses. I also instructed the subjects to try to keep their gaze inside of the range of what the hardware could capture. I was hesitant to instruct users to do this for fear that it would influence the results, but I had to be able to capture the data.
