Extraction of Precise Road and Lane Geometry from 3D Data (Masterthesis)


Khalid Bin Farhan

05.12.2014, 14:00, room 3945


The precise information about road and lane geometry is very important for many applications in the domain of Driver Assistance Systems. Various methods using Computer Vision techniques have been proposed and developed to extract the precise road and lane geometry.  However, most of these techniques consider an image from the camera as the only source of information.

In this thesis, an online system is developed that is capable of providing road and lane geometry information. The system not only uses the 2D images and 3D point clouds from the visual sensors, but also an open source map as source of information as well. This map provides the semantic information about the spatial environment around a vehicle and a searching template for the expected lane geometry which is combined with the data from visual sensors such that, precise geometry about road and lane can be obtained.  To determine the quality of our system, the ground truth data has been generated and appropriate metrics have been defined. In the prototypical implementation, the output of the system is visualized using Augmented Reality and the system has been integrated into the existing framework in a test vehicle.