TREND Automotive: A heart for pedestrians

| |
1 Star2 Stars3 Stars4 Stars5 Stars

Driver assistance systems are increasingly focusing on endangered pedestrians. These systems can see more and more like drivers but brake and maneuver with the quickness and precision of machines.

Pedestrians and cyclists are at a clear disadvantage on the road. Accident statistics reveal that 523 pedestrians died on German streets last year, 15.5 percent of all traffic deaths in Germany.

Though passive solutions such as pedestrian airbags or hoods that lift slightly during a collision with a pedestrian and “cushion” the impact can reduce the risk of injury, they are not enough. So in the future, driver assistance systems will improve the safety of “endangered” pedestrians.

Also because the requirements for crash test evaluations are becoming stricter. The safety organization Euro NCAP no longer just crashes cars into walls. It also awards its coveted stars for pedestrian protection and for vehicles equipped with active assistance systems.

Driver assistance for pedestrians

Driver assistance (Picture: Bosch).
Driver assistance systems detect obstacles that suddenly appear, such as pedestrians, and enhance – when necessary – the driver’s steering maneuvers to avoid them. (Picture: Bosch).

For example, Bosch is working on a new system that can help drivers with both braking and swerving before an impending collision with pedestrians. If braking alone is not enough to avoid a collision with a pedestrian who suddenly appears, the “electronic co-driver” instantly calculates the path for an evasive maneuver. As soon as the driver initiates the life-saving maneuver, the driving assistance helps with the steering. Research shows that 60 percent of collisions can be avoided when the driver reacts at least half a second before the collision.

A crucial component of the lifesaving setup is a stereo video camera from Bosch, of which production models are already in service. The camera, which is installed behind the windshield near the rear-view mirror, provides a three-dimensional image of the area in front of the car and detects pedestrians, oncoming traffic and obstacles on the road. On detection of a pedestrian, a computer in the trunk calculates the probability of a collision and a possible route for avoiding it – more than ten times per second. Easier said than done, as the algorithms have to use the image data to calculate where the pedestrian is expected to be a second in the future.

The city – beautiful but dangerous

The research initiative Ur:ban is also dedicated to pedestrian safety. Its goal is to improve driver assistance and traffic management systems.

As part of the initiative, Daimler researchers have made a significant advance. Using so-called scene labeling, a camera-based system classifies completely unknown situations automatically, recognizing all objects relevant for driver assistance – including bicyclists, pedestrians and even wheelchair users. The process involves distinguishing 25 different object classes such as vehicles, bicyclists, pedestrians, streets, sidewalks, buildings or trees in thousands of images of different German cities. When fed with such information, the system has learned to correctly and automatically classify completely unknown images and recognize them even when largely obscured or at great distances. The immense computing power is provided by computers wired similarly to the human brain in so-called deep neural networks.

With these capabilities, the system is very similar to human vision, which is also based on a complex neuronal system that links information from the individual sensory cells on the retina so that a person can recognize and distinguish a nearly unlimited number of objects. Scene labeling transforms the camera from a pure measuring system to a cognizant system as versatile as the eye-brain system.

Seeing like people but reacting like machines – some day that will be the solution for many problems, at least on the road.

Scene labeling. (Picture: Daimler).

Where is the pedestrian? With scene labeling, the camera-based assistance system is capable of recognizing even complex traffic situations. (Picture: Daimler).