Un séminaire de l'équipe A3SI du LIGM (unité mixte de recherche de l'Université Paris Est) a eu lieu le mercredi 5 décembre de 13h30 à 16h00 à ESIEE Paris.
Abstract: Self-driving cars use various sensors, to be tested on extensive test fields under location-, weather-, and traffic-specific conditions. One test field (N3T) is currently developing near Whangarei in New Zealand, in collaboration with the German Air- and Space Centre (DLR) and Auckland University of Technology (AUT). In the context of test field studies, the talk informs especially about computer-vision based components towards self-driving cars. Visual odometry supports exact geo-localisation of vehicles on the road, and in particular also accurate 3-dimensional roadside reconstruction, thus improving GPS/IMU-only based approaches.
Different camera configurations (mono, bi, or tri-nocular) contribute to the options for sensor configurations in self-driving cars. The talk discusses results for visual odometry and stixel calculations (stixels are an important intermediate result within a semantic segmentation framework) for evaluating different camera configurations. It is also demonstrated how visual odometry provides important information for improved road-surface distress analysis.
The talk reports about joint work with colleagues and students at AUT, DLR, and N3T.
Abstract: We survey some models of stochastic multi-agent interactions, involving simple ant-like a(ge)nts moving in grid or general planar graph environments, leading to interesting results concerning the average number of visits to various sites and to connections between Euclidean and discrete geometry.