F2008-08-148
Early Detection of Hazards in Driving Situations through Multi-Sensor Fusion
Road Accident Statistics show that driving can often be a dangerous activity. Statistics reveal that accidents are mostly caused by several circumstances. The majority of current Advanced Driver Assistance Systems (ADAS) focus only on single aspects of the driving situation, such as lane departure warning or collision warning. A new approach to address this problem is to consider the driving situation as a whole. As many single components as possible as well as spatial and semantic relations are included. In order to achieve this objective different kinds of sensors ought to be integrated and combined with techniques of sensor data fusion. The present proposal is based on this paradigm and sets the course for a sensor system for early detection of hazardous driving situations. The underlying sensor network does not only consist of conventional vehicle sensors such as engine sensors. Also advanced sensores like radars, laser scanners, cameras, high precise positioning systems and inertial sensors are integrated. The applied blackboard architecture and the modular structure of the system, enable the different parts of the system to easily exchange information and thus to extend their knowledge about the driving situation. Currently developed modules are principally concentrated on the analysis of the driving environment and can be divided into three groups:
(i) Infrastructure: How does the stationary environment look like? What kind of street is it? In which lane of the road are we driving? Are we approaching an intersection? It is attempted to provide answers for all these questions using laser scanner data, a lane detection system and conventional digital maps. (ii) Other traffic participants: Are there other vehicles around? What are they doing or intending to do? Are there other objects? For answering these questions, laser scanner data is fused with an image based object detection and classification. With it, the data integrity of both sensors can be improved. (iii) Driver actions: Is the driver driving properly according to the road, light and weather conditions? Is there a hazard of potential collision with another vehicle? Is the driving speed and maneuver allowed in this situation? These questions can be answered by combining the obtained information of (i) and (ii) and adding some new information about the driving maneuver.
The output of each module is drawn into a car centered coordinate system and the higher level information is fused in a danger map. The advantage of this method is both that potential dangers can be detected and the origin of a hazard can be reconstructed independently of the underlying sensors and processing modules. Based on different sensors and fusion algorithms the proposed system allows easily the extension through addition of new modules.
Poster presentation: Vehicle safety