Sensordatenfusion und Bildverarbeitung zur Objekt- und Gefahrenerkennung

Kurzfassung

The present work deals with automatic detection and tracking of objects in driving situations as well as derivation of potential hazards. To do this, the data of a laser scanner and a camera is processed and fused. The work provides new methods in the area of immediate environment detection and modeling. Thus, it creates a basis for innovative driver assistance and automation systems. The aim of such systems is to improve driving safety, traffic efficiency and driving comfort. The methods introduced in this work can be classified into different abstraction levels: At sensor data level, the data is prepared and reduced. In this work, the focus is especially set on the detection of driving oscillations from camera images and on the detection of the driving corridor from the data of different sensors, used later as the primary area of interest. At object level the central data fusion is done. High reliability, availability and sensor independency are achieved by choosing a competitive object fusion approach. As an input of the data fusion, object observations from camera and laser scanner data are extracted. These are then fused at the aim of object detection and tracking, where aspects such as robustness against manoeuvring objects, measurement outliers, split and merge effects, as well as partial object observability are addressed. At application level, early detection of potential hazards is addressed. A statistical approach has been chosen and developed, in which hazards are handled as atypical situations. This general and expandable approach is exemplarily shown based on the detected object data. The presented strategies and methods have been developed systematically, implemented in a modular prototype and tested with simulated and real data. The test results of the data fusion system show a win in data quality and robustness, with which an improvement of driver assistance and automation systems can be reached.