Collective Perception in Vehicular Ad-hoc Networks

Abstract

In combination with the current developments in the area of automatically driving vehicles, the introduction of inter-vehicle communication plays a crucial role for realising the long-term objective of what is known as cooperative driving. A cornerstone for the expansion of automated vehicles is their thorough understanding of the current driving environment. For this purpose, each vehicle generates an environment model containing information about other perceived traffic participants and objects. Local perception sensors are important data providers for this model, as they contribute implicit knowledge about the environment. In combination with a direct communication link between traffic participants, explicit knowledge can be added to the environment model as well. The key concept developed within this thesis is called Collective Perception: it focuses on sharing data gathered by local perception sensors of one vehicle with other traffic participants by means of inter-vehicle communication. As a result of this concept, future applications relying on a comprehensive understanding of the current driving environment are made feasible. The analyses presented in this thesis employ a vehicular ad-hoc network (VANET) based on the standardised framework of the European IEEE 802.11p-based ITS G5 protocol stack for inter-vehicle communication. The effectiveness of the technology relies on an existing communication link between a sufficient number of communication partners - the critical mass. The expansion of inter-vehicle communication, however, can be supported by capacitating indirect effects. Collective Perception is one representative of these effects, as the information density within the network between the vehicles is increased, even at low market penetration rates. At the core of Collective Perception stands the introduction of a message format which serves as a vehicle for the exchange of sensor data within a VANET. The development of the message is influenced by two perspectives: First, the vehicle perspective affects the relevant contents of the message required by data-fusion processes and application algorithms. Second, from the network perspective, constraints resulting from the network stack and effects caused by congestion control mechanisms have to be considered. This thesis addresses both perspectives to develop a holistic concept for exchanging sensor data within a VANET.