Tracking Theory for Preventive Safety Systems
Licentiate thesis, 2008
Preventive safety systems rely on accurate information from a sensing system regarding the current traffic situation to make decision whether to inform, warn or intervene to avoid an impeding collision or to mitigate its consequences. This thesis mainly considers tracking algorithms to enhance these systems through making the best use of the information supplied by on-vehicle sensors, such as radar and vision sensors.
In this thesis we propose a sensor data fusion framework to support the utilization of information about the current traffic scene provided by several sensors. This framework is based on a modular approach, enabling the exchange of sensors and tracking components with relative ease. Furthermore, it considers many of the practical aspects that arise when designing a real-time fusion system. For this type of system to fully utilize the information provided by the sensors, two kinds of probabilistic models are needed, i.e. motion models and sensor models. Here, we present contributions to the design of both of these.
Research show that radar signals are more likely to be reflected in certain specific structures on a vehicle, such as headlight cavities and wheel housings. By associating radar detections to these specific structures accurate information can be extracted regarding , e.g vehicle position and orientation. We present a radar sensor model family that is able to exploit this added information to improve the tracking performance. Furthermore, we propose a vehicle motion model framework to accurately describe the motion of a vehicle by including the expected driver control input into the model. The proposed framework not only enables more accurate predictions but also offers a formal treatment of the model uncertainties. The latter is a very useful property when applying the framework in a tracking algorithm.
Additionally, aspects of integrating several preventive safety systems in a vehicle are investigated in terms of not overloading the driver with too much information and warnings. To reduce the stress for the driver we propose arbitration and prioritization schemes for those situations when multiple warning systems require the attention of the driver.
sensor data fusion
active safety systems
human machine interface
preventive safety systems