Speaker: Dr John McDonald, Maynooth Unviersity Department of Computer Science.
Title: "Visual SLAM: from Sparse Mapping to Dense Perception".
Abstract: Visual Simultaneous Localisation and Mappping (vSLAM) refers to the problem of computing a model of an environment froma moving camera, whilst simultaneously estimating the motion of the camera relative to that model. Solutions to this problem have been key to recent developments in areas including autonomous mobile robotics and markless augmented reality (AR). In this talk I will provide an overview of the visual SLAM problem in the context of a number of vSLAM systems that my research group has developed. The first system employs a feature-based approach for multi-session visual mapping where multiple separate mapping sessions can be efficiently combined into a single globally consistent model of the environment. The second system utilises a more recent featureless paradigm to allow the computation of mesh based models over extended scales. Finally I will show some initial results of our current research on (i) the development of collaborative dense SLAM, and, (ii) the incorporation of machine learning for 3D in-painting within the SLAM pipeline.