Pick it with your eyes – an eye tracking experience
Automatic extraction of features from scientific images is of extreme relevance especially in the field of space exploration. In fact, in space exploration a huge number of images is taken, which can't be analyzed in realtime due
to distance and therefore delay in the trasmission; at the same time, it's unpractical to transfer all the images to hearth for an offline analysis, due to memory constraints.
It would be ideal if the robot could evaluate in real-time the scientific content of a picture, assigning it an index to set its relevance. Many works have dealt so far with the definition of such an index. In particular, a lot of research in ongoing on developing machine learning algorithms able to autonomously identify interesting pictures; the preparation of the training set can be performed by manually labeling interesting items in pictures. The goal of this project is instead to label the pictures of the training set not manually but using human gaze data collected through an eye tracker. This way, the machine will learn to look at the scene the same way as a human would do.
An Eye Tracker is a device capable of tracking the direction of the human gaze while looking to the screen of the computer.
Eye Tracking on the Open Day 2015On the ESTEC Open Day 2015 almost hundret participants took part in the eye tracking experiment. Visitors to the Open Day were curious about where they look at and could inspect their personal results immediately after the experiment.
The experiment conducted followed a strictly timed schedule. Every person looked at five pictures for seven seconds each. Before each picture a circle was shown in the center of the screen for two seconds to ensure that participants are looking at the center of the screen when a new picture is shown. The pictures from the surface of Mars were selected randomly out of a set of 48 pictures. The end result of the day are seventy data sets have a high enough accuracy for further evaluation.
The following images show the first evaluation of the recorded data. Each picture displays a so called heat map where the color changes depending on how long people looked at one spot. Additionally the pictures show areas in red which have been annotated by geologists as interesting features of the image.