Environment modelling by sensor data fusion in dynamic grid maps
The environment modelling enables the autonomous vehicle to do a collision-free movement in its environment. For this purpose, the free space as well as static and dynamic obstacles must be calculated. Particularly the direction of movement of the dynamic obstacles is of interest, as they directly impact on the free space. This task of environment modelling is performed by the dynamic grid map.
In the approach of the dynamic grid map, the environment is disrected by dividing it into cells. A cell contains information about the occupancy probability and the direction of movement of dynamic objects. The information for calculating the content of a cell is combined by various sensors, such as laser scanners or radars. To combine the information from different sources, a sensor fusion takes place in the dynamic raster map. Probabilistic methods and approaches in the field of Dempster-Shafer theory, particle filters and random finite sets are used for sensor fusion and calculation of the dynamic grid map.
This project researches a more robust estimation of objects in the environment. Methods for object identification and fusion with other sources of information are investigated and further developed.