PixelDust: Supporting Dynamic Area of Interest Tagging in Programming Studies with Eye Tracking
Eye tracking studies are valuable for evaluating program- ming environments, but annotating what the programmer is looking at in a dynamic environment can be repetitive, time- consuming, and error prone. Through a participatory design exercise with two eye tracking researchers, I identified three significant challenges: search, extraction of code, and annotat- ing transient objects. By applying computer vision algorithms to video traces, I developed a mixed-initiative system called PIXELDUST, which allows the researcher to train a system to recognize different objects on a screen. My preliminary results demonstrate the versatility of the approach; for ex- ample, the system can recognize return statements, method signatures, tool tips, and dialog boxes.