Perception and Modeling
The IV group investigates and developes state-of-the-art techniques for vehicle perception, and modeling the vehicle’s environment, containing static infrastructure and dynamic objects. Important classes of dynamic objects are pedestrians and (motor)cyclists, the so called Vulnerable Road Users (VRUs), but of course other vehicles, buses, trams and trucks are detected and tracked too. Perception of the environment can be done with a multitude of sensors, including radar, lidar, and cameras. Our group’s demonstrator vehicle for instance utilizes a stereo-camera setup to create a 3D representation of the environment, and detect its occupants, as seen in the image on the right.
By understanding where objects are around us, and what kind of objects these are, an intelligent vehicle can also predict future traffic situations, and assess the risks and benefits of different manouvres. Modeling behavior of traffic participants is therefore an important research topic in our group.
A key necessity for safe and comfortable automated driving is predicting possible future outcomes of the surrounding traffic situation. Therefore, perception not only concerns determining what traffic participants are around the vehicle, and where they are, but also what their behaviors is, and what they will be doing next. Vulnerable road users such as pedestrians can quickly change their path, for instance when crossing the road to stopping at the curbside. To make better predictions, the automated system needs to consider factors that affect the pedestrian’s decisions. Our research considers different types of contextual factors, based on the road layout, interaction between the road users, and the pedestrians awareness of the approaching vehicle (has the pedestrian seen the approaching vehicle?). Including all these factors in a probabilistic framework leads to larger prediction horizons, resulting in earlier warnings without raising more false alarms. See also
Kooij, J. F. P., Schneider, N., Flohr, F., & Gavrila, D. M. (2014, September). Context-based pedestrian path prediction. In European Conference on Computer Vision (pp. 618-633). Springer.