Deep learning used to recognise eye gaze

German researchers have used deep learning to transform cameras into devices that can detect whether a person or group of people are making eye contact with an object.

The technology could potentially be used to judge, for example, whether people are looking at a poster in a crowded pedestrian area.

The research team have tested the camera in two different situations. They have mounted it on a target object in a workspace and also captured a first-person perspective using a body-mounted camera.

In order for the device to work, the camera must be placed close to the target object.

The gaze direction of individuals is clustered and used to estimate the likely location of the object. This creates an object-specific eye contact detector.

Estimation of the object location and dimensions improves over time as more data is accrued.

Researchers report that the device works regardless of variation in the number of people, lighting conditions, camera position and the dimensions of the target object.

The research is being conducted as part of the Perceptual User Interfaces group, which is a collaboration between Saarland University and the Max Planck Institute for Informatics.

Video credit: Perceptual User Interfaces Group