Csaba BENEDEK

Abstract:

In the past decade we have witnessed an explosion of new technologies for acquisition and understanding of environmental information. 3D vision systems of self-driving vehicles can be used for -apart from safe navigation- real time mapping of the environment, detecting and analyzing static (traffic signs, power lines, vegetation, street furniture), and dynamic (traffic flow, crowd gathering, unusual events) scene elements. On the other hand, new generation geo-information systems (GIS) store extremely detailed 3D maps about the cities, consisting of dense 3D point clouds, registered camera images and semantic metadata.

In this talk, I present new techniques to facilitate the joint exploitation of the measurements of car mounted online sensing platforms, and offline 3D environmental data obtained by Mobile Laser Scanning (MLS) technology in urban environment.

• First RangeMRF, a Lidar based real time and accurate self-localization and change detection approach is presented for self-driving vehicles, using as reference high resolution 3D point cloud maps of the environment obtained through Mobile Laser Scanning (MLS).
• Second, we propose an end-to-end, fully automatic, online camera- Lidar calibration approach, for application in self-driving vehicle navigation.
• Finally, ChangeGAN, a novel deep neural network-based change detection approach is introduced, which can robustly extract changes between sparse and weakly registered point clouds obtained in a complex street-level environment, tolerating up to 1m translation and 10 degrees rotation misalignment between the corresponding 3D point cloud frames.