Benefits of using explicit ground-plane information for grid-based urban environment modeling

Future advanced driver assistant systems and automated driving put high demands on the environmental perception especially in urban environments. Major tasks are the perception of static and dynamic elements along with the drivable area and road structures. Although these tasks can be done without an explicit representation of the ground surface, evaluations with real-world sensor data have shown that many failures in sensor data interpretation result from inappropriate ground point filtering, e.g. false negative or false positive detection of objects. That means, the ground plane estimation becomes a key-task in urban environment modeling assuming a reliable perception of the local vehicle's surroundings is required. In this paper we thus present the enhancements of the environment modeling that can be derived from an explicit representation of the ground surface. We propose the combination of channel-wise ground classification with a grid-based representation of the ground height to cope with spatial false measurements and exploit interchannel dependency. Partial occlusions by other road users are handled in an efficient way. Integration of the ground data is shown by a pitch angle estimation and a curb detection module. The algorithms run in real-time on a standard PC and are evaluated with real sensor data.