Information fusion for cognitive automobiles

During the last decades, the steady increase in road transportation of people and goods has had a significant negative impact on traffic efficiency, safety, and the environment. The problems arising from the growing automobile usage have motivated researchers from all over the world to develop solutions – many of them based on information and communications technology – to reduce traffic congestion, air pollution, and road traffic fatalities. Consequently, automobiles are increasingly being equipped with a number of different sensors and will likely be capable of automatically communicating and exchanging information with other cars and with a technical environment in the future. Moreover, a high amount of prior knowledge, especially about the road network and the traffic infrastructure, will be carried onboard. Based on more powerful communications protocols and devices, road users will be capable of accessing large pieces of information from different sources, e.g., through the recent IEEE 802.11p standard [1]. However, the question arises, how this multitude of available data, information and knowledge from both onboard and external sources can be fused and utilized to enable cars to develop cognitive capabilities and act intelligently – and thus to make them safer, more efficient, comfortable, and environmentally sound. ‘Information Fusion’ has already devoted a recent special issue to Intelligent Transportation Systems (ITS), in which a broad overview of both the state-of-the-art of the ITS infrastructure and the upcoming ITS applications was offered [2]. For this reason, fusion technologies explicitly involving infrastructure elements are beyond the scope of this special issue. The focus is rather on onboard fusion technologies enabling cognitive functions, such as assisting the driver or even making him superfluous, as is the case in autonomous cars. To implement such advanced cognitive functions, perception and understanding of the environment are crucial prerequisites. This encompasses several basic skills. First, the automobile needs to sense its environment in a comprehensive way based on complementary sensors. Then, the acquired sensor data has to be transformed into a symbolic representation, in which the objects in a scene along with their types and relationships are identified. Finally, semantic information has to be extracted, such as the ‘meaning’ of different object constellations or the intentions of other road users. Due to the criticality of this field of application, all algorithms should operate in real-time and behave robustly in the presence of noise, variable lighting or weather conditions, occlusions, dynamic backgrounds, and other influencing factors. As a response to the call for papers that announced this special issue, a large number of original technical contributions covering new methods and algorithms were submitted by teams from all over the world. After a careful review process, five of them were selected to assemble this special issue. An additional survey article is