Japanese and German researchers have just created extremely smart cameras that can tell exactly where people are looking. All of these new cameras use the latest advances in deep learning technology.
The researchers have successfully tested these cameras in two locations. First, they placed a camera on a designated object in an office. Second, they used a body-mounted camera to get a first-person perspective.
As of today, the camera can only detect people's gaze if it's placed near a specified object. These cameras are able to pick up the general gaze of a crowd of people and guess whether or not they are looking at the targeted object. Of course, the more data the cameras have, the better the results from the test will be.
This technology has great applications in the field of marketing science. For example, researchers can place a camera near advertisements in a metro station and gauge how many people look directly at the signs and for how long.
As far as the researchers can tell, the cameras work well under any lighting conditions and from any position. These cameras can also collect data from extremely large quantities of people.
For those who aren't aware, deep learning refers to a special branch of machine learning that helps robots perform complicated tasks. Technicians based their science on an intricate study of how the brain sends electrical signals and applied these findings to machine programming. Besides visual learning, deep learning can be used in audio programming and speech recognition.
The principal researchers on this camera include Xucong Zhang, Andreas Bulling, and Yusuke Sugano. The first two technicians are a part of partnership between the Max Planck Institute for Informatics and Saarland University called the Perceptual User Interfaces group. Mr. Sugano studies at the Graduate School of Information Science and Technology at Osaka University.
Anyone can check out these researchers' studies in a short video clip on YouTube. Perceptual User Interfaces posted this video under the title "Everyday Eye Contact Detection Using Unsupervised Gaze Target Discovery (UIST'17)."