We started of the second week with our first coach meeting with Willemijn Elkhuizen. We quickly found out that our understanding of the assignment did not line up with what Willemijn had planned for us. Instead of doing research on the difference in perception of a painting between experts and regular people, we have to build a working research setup, as Willemijn did not have one yet.
The current eye-tracking system only works with a monitor. However, a painting is a physical object. Our job for this project, is to make the available eye-tracking software work with a physical object, like a painting. An additional challenge was added, because the perception of the painting has to be adjustable, because the gloss really stands out when the light direction changes.
On Tuesday we divided the tasks and got to work. Celine analysed the software, Jerome started writing an Arduino code for the motor and Ronja thought of ways to move the painting/light in the most effective way. As for changing the perception, we decided to keep the painting stationery and build a track on which a lamp can moved in a circular motion around the painting. This way, we can change the perception without having to move the painting or the test subject. By the end of the day, we had some ideas on how to make the tracking software work with a physical object.
Thursday, we tried to make a working prototype of our research setup. Our goal was to make a working setup with a stationary painting. We decided to add the changing perception later.
Our research set up consisted of five components:
- our laptop with the tracking software
- the Tobii eye tracker
- a beamer
- a painting with a white framework
- a camera
The beamer was used to project the calibration process. This process works as follows: the beamer projects a dot on the screen, which moves to certain points. The test subject has to follow this dot with their eyes, which are tracked by the tracking device. Then we can check on the laptop whether the tracking device tracks the eye movements accurately enough.
We could have drawn our own calibration grid. However, using a projection is way easier and more exact, because a slight change in distance or height, would not require a new grid. We would simply have to change the parameters in the tracking software.
The camera was used to record the test. This camera footage was used as an underlay for the test results. For this, the camera footage also had to be calibrated with the entire setup.
Because of the amount of different components that had to be fine-tuned with each other, and because the was no manual on how we could make this calibration work, it took us an entire day of fine-tuning and changing things to make it work correctly. Everytime we changed something, we did a test to check the accuracy. We gave the test subject the task to look at the center of the painting and then at every corner. We analyzed the test results to see whether they were accurate enough (see videos below). Note, we used a crappy webcam for our prototype. The real setup includes a high quality camera.
Luckily, our hard work paid off, as we had a working research setup prototype at the end of the day. Our main calibration mistakes were the following:
- The angle between the eye tracking device and the test subject’s eyes was too big. This angle may not be bigger than 17.5 degrees.
- In the calibration process with the camera, we selected the painting as the ‘active area’. However, the screen projection of the calibration process should have been the ‘active area’.
Finally, we decided to add one last component: a head support so all test subjects view the painting from the same point of view. This means we don’t have to change the height of the eye tracking device every time we use a different test subject.
Now we knew how to make the setup work, we started designing frames for the components. By using frames with fixed ‘ideal’ dimensions, we reduce the amount of measuring and calibration work we have to do for each test. We started building the frames on Friday. Next week, we will continue building the frames and start building and testing our initial version of the research setup.