This article was originally published at NVIDIA's blog. It is reprinted here with the permission of NVIDIA.
Today’s crowded urban centers are, more than ever, a mine field for drivers. It’s not just that there are more pedestrians on the streets; many of them are staring at or talking on their mobile devices as they walk, rendering them oblivious to their surroundings.
Thanks to GPUs, researchers at Japanese automobile parts-maker Denso are closer than ever to equipping cars with the ability to maneuver on their own to avoid pedestrians, Ikuro Sato, senior engineer in the research and development group at Denso’s IT laboratory, told attendees during a session at NVIDIA’s GPU Technology Conference.
By taking decision-making in perilous situations out of the hands of human drivers, Sato and his team hope to make the road safer for both motorists and pedestrians, and refine an important element of a future in which cars will drive themselves.
Sato emphasized what a complicated problem it is to solve, noting that distinguishing between pedestrians moving in different directions, at different points of the process of crossing the street, and possessing varying levels of awareness, is an enormous challenge.
In fact, Sato said, current technologies limit the ability of vehicle systems to simple identification of the presence of pedestrians and basic maneuvers such as slowing and stopping. What he and his team are attempting to do is extract richer information using image recognition technologies so the autonomous maneuvers of a vehicle match those of the most skilled human drivers.
Previously available technologies, said Sato, have been unable to:
- Build on the binary classifications they perform by adding needed metadata on pedestrian characteristics;
- Achieve the levels of so-called “visual feature design” necessary to distinguish, say, between a person walking empty-handed and one using a smart phone; and
- Perform real-time processing.
