Abstract
Keywords: pedestrian detection, pedestrian recognition, impact prediction, advanced driver assistance systems, night vision, system evaluation, collision mitigation.
Passenger cars are still contributing most to the statistics of fatal traffic accidents with pedestrians worldwide. While car passengers could increasingly profit from active and passive safety developments in last years, every eighth person who dies on roads today is a pedestrian.
Car manufacturers and authorities in US and EU discuss countermeasures to efficiently improve the safety of pedestrians in traffic environments. In the next car generations, passive and contact-active safety systems will thus aim to reduce the risk of serious injuries in the event of a collision with a pedestrian. But key technology solutions have furthermore to be investigated to prevent accidents involving pedestrians and cars. The trend goes to pre-crash safety systems.
In this context, more and more attention is given to image processing approaches supported by optical sensors. While the first camera applications were dedicated to provide driving relevant information to the driver, cameras tend to establish themselves in the automotive industry as a new sensor family for computational vision. This evolution could enable revolutionary improvements in pre-crash safety systems, which could be deployed as the most efficient pedestrian protection.
Two sorts of applications can be distinguished: the one trying to inform or warn the driver as early as possible, so that he has enough time to react; the other operating shortly before the crash when there is no time any more to involve the driver but rather activate autonomous reaction like automatic breaking or triggering pedestrian protection. According to the type of application, the BMW Group selects the most suitable camera technology to satisfy the challenges.
Because of the significant high risk of pedestrian accidents while driving at night camerabased systems dedicated to the far zone are developed. An already existing far zone application is the BMW Night Vision system based on Far-Infra-Red (FIR). The BMW Night Vision system uses the FIR electromagnetic spectrum to provide a clear-sighted view under almost all environmental conditions at night. This offers the customer more safety and an increased sovereignty.
The system depicts the area in front of the car up to 300m without the need of any active light source just using the heat radiation given by every object. Especially pedestrians with a typical temperature of 309 Kelvin are well distinguishable from the surrounding thanks to system optimization for this temperature range.
In comparison to Visible or Near-Infra-Red (NIR) systems there are no malfunctions by glare effects due to vehicles with high beams or NIR sources. The image is shown in the central information display, which assists the drivers to prevent from imminent dangers.
In the case of camera-based pedestrian detection dedicated to pre-crash applications in near zone, the challenge is two-fold. In addition to the realization of a pedestrian detection with high-fault safe stage, an impact prediction application has to be achieved based on the exact localization of the pedestrian, his speed and his moving direction.
Different methods are investigated at the BMW Group for the evaluation of camera-based pedestrian detection systems. More specifically, their 3D performance is under analysis. One approach consists in referencing videos recorded under application specific test conditions using differential GPS systems, so that the relative positions of the pedestrian to the vehicle can be calculated and compared in real time.