Optical image recognition for intelligent vehicles
At a road junction with signs, traffic lights, cyclists and pushchairs, a sensor has to perceive its environment in a particularly discriminating way. This is why Mercedes-Benz is conducting research into video-based driver assistance systems which effectively teach the car to "see".
Information provided by video cameras offers the closest approximation to human perception and forms the basis for a comprehensive recognition system. The researchers developing this image-based approach to environmental perception are using two methods: optical flow and stereo image technology.
In the optical flow method, the camera behind the windscreen picks up the entire traffic situation in individual images. Special, efficiently programmed software calculates changes in the position of each key point from one image to the next.
Stereo image technology works with two cameras in a stereo arrangement. By linking the two methods using intelligent algorithms, the computer is able to identify stationary and moving objects in real time – even independently of the environment. Furthermore, the computer can also determine in real time the speed at which the vehicle is approaching the moving objects.
Electronics can now safeguard against every driver's worst nightmare: a child suddenly appearing between parked cars. Perception of the child's head alone is sufficient for a computer response. Within 80 milliseconds, the system registers the danger and can trigger the necessary braking. With a human driver, by comparison, the visual information would have merely travelled from the eye to the first vision centres in the brain over the same period of time.