Gabrielle Flood: Motion Maps with Statistical Deformations

With the fast development of autonomous vehicles and agents that move in the world without human supervision, the need for good localisation methods is high. However, the agent cannot determine its position without having something to relate to, hence, there is also an increased demand for accurate 3D maps. An important issue connected to this is how to match and merge individual local maps into one global map.

In this talk we will discuss how to do efficient, flexible and robust map matching and merging. We mainly focus on maps created from image data, but the methods are applicable to sparse 3D maps coming from different sensor modalities. Merging of maps can be advantageous if there are several map representations of the same environment, e.g. in collaborative SLAM, or if there is a need for adding new information to an already existing map. We will look at ways to do this which utilises a compact approximation of the residuals and still allows for deformations in the original maps.

Gabrielle Flood is a postdoctoral researcher at the Division for Computer Vision and Machine Learning at the Centre for Mathematical Sciences, Lund University, where she also defended her Ph.D. in applied mathematics in 2021. The topic of her thesis was “Mapping and Merging Using Sound and Vision - Automatic Calibration and Map Fusion with Statistical Deformations”. Prior to this, she received a M.Sc. in applied mathematics in 2016, also from the Faculty of Engineering at Lund University. Gabrielles research interests concern image analysis in general and geometric problems in computer vision in particular. Currently, her research is mainly focusing on positioning and robust estimation of 3D reconstructions from different sensor modalities - such as images and audio - and efficient merging of such reconstructions.