AASS Seminar - Exploiting Structural Knowledge for Visual Place Recognition in Changing Environments

The research centre AASS arranges a seminar with Stefan Schubert, T.U. Chemnitz.

Abstract

Visual place recognition is the task of finding query images in a database of previously recorded images. It is required for tasks like loop closure detection in SLAM and candidate selection for global localization. While there are existing well-performing methods for place recognition under constant environmental conditions, visual place recognition under changing environmental conditions due to long-term operation is still challenging and part of active research. Place recognition as an embedding of image retrieval in the context of mobile robotics allows the exploitation of additional structural knowledge and information like spatio-temporal image sequences or intra-database similarities. The talk will start with an introduction to visual place recognition in changing environments and algorithmic elements of a basic place recognition pipeline. Subsequently, methods for the extension of such pipelines are discussed that exploit structural knowledge for performance improvements. Finally, potential pitfalls of such methods are presented which should be known for a successful application on new data.

Go to Zoom

Bio

Stefan (https://www.tu-chemnitz.de/etit/proaut/stefanSchubert) is a final year PhD student at the Chemnitz University of Technology, Germany, and received his master's degree in electrical engineering in 2014. He was selected as a 2019 RSS Pioneer and awarded with two best paper awards during the UK-hosted international TAROS conference 2016 and 2017. His research interests include visual place recognition in changing environments, high-dimensional computing, camera-based localization and deep learning, particularly for application in autonomous mobile robotics. His work is driven by techniques from computer vision, machine learning, graph-based optimization, high-dimensional computing, and bio-inspired methods.