Predicting pedestrian behavior

Is it possible to predict a pedestrian’s behavior? This is a serious question that autonomous vehicles need to ask, and answer, as they maneuver busy streets.

It’s a question that, concerning pedestrians, we ask ourselves routinely as we drive — whether the person standing on the corner is about to step into the street, run across it, or just stand still as we pass.

We learn to predict behaviors by processing a series of cues: how people are standing, in which direction they’ve turned their heads, the expressions on their faces, and so on. Although we take this ability for granted, we’ve learned it over many hours of observation and mental processing.

Using machine learning and Micron’s deep learning accelerators, Continental’s advanced driver-assistance systems (ADAS) are developing artificial intelligence capabilities that enable autonomous vehicles to do the same (read the press release here). By learning from many hours of video captured by vehicle cameras, these computers are not only identifying pedestrians on the roadsides, crossing the street, and in other places close to the vehicle, but they are predicting what they’re about to do next and proactively taking action.

Watch these short videos to see these amazing capabilities in action — and learn more about the specially engineered Micron deep learning accelerators that make it all happen. And hear from the experts about how soon we can really expect to see fully self-driving cars on the road. (Hint: It’s sooner than you might think.)