Detailed on the startup’s official site, the development team behind PredictGaze are attempting to recognize specific actions utilizing front-facing cameras on various types of consumers electronics. For instance, a camera that’s watching someone while they watch television will be able to spot when a person walks out of the room to grab a snack from the kitchen or grab a phone call. The PredictGaze software automatically pauses the content playing on the television screen in addition to starting it back up again once the person has walked back into the room and sat down on the couch. This would be ideal for darkened home theaters rather than attempting to fumble around for a remote lost in the couch cushions.
Demonstrated within the demo, the system is able to identify multiple people in the room and can tell if they aren’t fixated on the television screen. According to the developers, the software can recognize actions up to four feet away using a standard VGA camera and up to twelve feet away using a 4-megapixel camera.
For the purposes of the demonstration, the team was using the rear-facing camera on an Apple iPad in order to watch users watching television. While similar to Microsoft’s Kinect system for the Xbox 360, CEO Saurav Kumar and co-founder Ketan Banjara are attempting to lower the cost of entry with less expensive hardware.
Regarding security, all image processing is being done locally rather than outputting the images or the feed to a server for processing. Hypothetically, a consumer electronics manufacturer could add this technology to a device without the fear of a consumer backlash regarding their privacy. In addition, devices using the PredictGaze system only store data points rather than specific images or video.
In addition to the TV feature, there are additional functions of PredictGaze that can be used in a variety of situations. In another product demo, the user is listening to music on their laptop while working at a kitchen table. When someone calls on the phone, the user simply places a finger up to their mouth in a “Shhh” gesture to silence the music.
Utilizing gaze tracking, the PredictGaze system can watch the user’s eyes and can automatically scroll up or down a Web page within a browser on a tablet. Another application of PredictGaze could be using facial recognition to automatically darken a tablet’s screen when the user passes the tablet to someone else when the tablet is logged into something private like a social network.
Beyond these initial scenarios, the group behind PredictGaze is also looking into age, emotion and gender detection. For instance, the system could automatically pause a movie not suitable for children and darken the screen if the camera on the television detects that a young child has entered the room. Using emotion detection, the system could tell what portions of a game on a tablet made you smile and what portions made you scowl in concentration. At this point, all this data could be collected and anonymized before being supplied to the content creator for analysis. However, it’s highly likely that the user would want the option of opting into that type of information sharing in order to preserve privacy.