The complexities of urban traffic environments pose challenges that exceed the capabilities of currently available detection and assessment technology by a considerable margin. Effective accident prevention in urban areas requires a reliable and comprehensive “picture” of the environment and a correct interpretation of difficult situations involving multiple protagonists and complex boundary conditions. Thus, in addition to a high-performance, all-around view, effective modeling of static and dynamic vehicle environments is needed that can serve as a basis for multiple applications.
This task is best accomplished in a horizontal sub-project establishing all-around detection and precise modeling as a basis for the applications of the UR:BAN project “Cognitive Assistance”. The goal is to implement generic representations of the traffic situation in which the intelligent sensors are decoupled from the specific application. In this way, a multitude of sensor variants and configurations can be tested, providing a robust technological platform and basis for future driver assistance systems.
Different spatial and temporal perceptual representations
In this cross-sectional sub-project, all-around, 360-degree environmental perception and detection models are being developed for the first time on a stand-alone basis. These models will allow application independent environmental models that can integrate data not only from currently implemented sensors, but also from future digital maps and information sources.
Complex situations in urban areas are often characterized by the presence of relevant objects that are partially occluded. Their detection and tracking for determination of a precise state of motion constitute an unsolved problem up to now. Most heuristics used up to now do not meet requirements for safety critical systems.
The sub-project “Measurement and Modeling of the Environment” is expected to provide important progress on this point by improved fusion algorithms for heterogeneous data sources. Each level in the data fusion sequence – combination of data from different sensors, interpretation, and prediction – requires a reliable estimate of the confidence and plausibility associated with the sources of information at that level. To this end, a key aspect being addressed in the sub-project is a systematic treatment and assessment of uncertainties in the signal processing sequence, ranging from raw data to representation.