On Monday and Tuesday (March 31 and April 1) John Noble from SeeingMachines (all the way from Australia) gave us a training on how to get the most out of our eye-tracker, that his company produced. We already had experience with the system since it arrived in January. These sessions helped us a lot to improve our camera setup, our results, to hunt down problems and to get some very useful tips and tricks from the first hand.

From left to right: Oszkar Palinko, Alex Shyrokov, John Noble (in front of the computer), Jon Oppelaar, Zeljko Medenica

For starts, John gave us an introductory presentation on the eye-tracker and its theoretical background. Then we discussed camera and IR pod placement, the very important process of camera calibration, driver face model creation and face feature selection in faceLAB.

Feature selection: always use the least changeable points on one’s face

In the remaining part of the first day we further investigated possible locations for the cameras and IR pods. We asked John a few million questions about this. After all, if we don’t get the cameras where we need to, all the subsequent steps would be affected.

On the next day, we started off with adjusting the times on all the computers using NTP. It is important to have them synchronized in order to be able to compare results from different systems (eye-tracker, simulator, video recorder, etc). After this, we set up and tested the SceneCamera, that allows us to record the scene appearing in front of the driver of the simulator and overlay the gaze intersetction on this video. Then, we tested all the possible camera/IR pod setups in the car. Here you can see John using a ruler to make the cameras co-planar, symmetric and of proper tilt. It is a wonder, how a simple ruler can solve that many things when it’s in the hands of an expert:

John and ruler vs. eye-tracker

In the afternoon of the second day, we performed a test experiment under John’s supervision. We discussed the best methods to visualize and analyze the immense amount of data collected by the eye-tracker. We also troubleshooted the video overlaying system and discussed options for placing the system into an actual test vehicle.

I was thrilled by how much information we managed to acquire from this visit. Using e-mails it would probably take months to find out about all this. It helped a lot, that we explored the system in the months before the training. This way we could focus on problems and finesses. Now we have the challenging task to organize and put down ‘on paper’ everything we learned to conserve this knowledge for future grad students.

John, thank you for having the patience to listen to all of our (sometimes overly complicated) questions and answer them exhaustively. I hope we were not too bothersome. We wish you all the best on your further journey through the USA and beyond!

Oszkar Palinko

Subscribe to the comments through RSS Feed