Autonomous navigation in a known environment is a core skill of autonomous mobile robots and vehicles. The navigation skill requires that the robot is able to recognize different places on the map when it revisits them repeatedly over time. In the simplest form, the map is stored as images or depth maps associated with location tags. Observed images, and queries, are matched to the map, gallery, and the location of the best match is retrieved as the location estimate. This vision problem is known as visual place recognition.
Mainstream methods and approaches are based on RGB images and utilize recent deep learning techniques. In robotics, the depth sensors, such as stereo, ToF, LiDAR and RADAR, are popular as they are more robust to visual changes such as illumination variation (time-of-day) and weather conditions. This thesis particularly focuses on LiDAR-based place recognition.
This PhD thesis has been made in Tampere University’s Doctoral School of Industry Innovations (DSII) and was sponsored by Sandvik Oy. More information about the DSII.
Public defence on Friday 24 March
The doctoral dissertation of MSc Jukka Peltomäki in the field of Computer Vision and Machine Learning titled LiDAR Place Recognition with Image Retrieval will be publicly examined in the Faculty of Information Technology and Communication Sciences at Tampere University at noon on 24th March 2023, in the Hervanta Campus of Tampere University, Tietotalo building, lecture hall TB109. The Opponents will be Prof. Tapio Seppänen from the University of Oulu, Finland, and Prof. Heikki Kälviäinen from the LUT University, Finland. The Custos will be Professor Joni Kämäräinen from the Faculty of Information Technology and Communication Sciences at Tampere University.
The dissertation is available online.
Photo: Jukka Peltomäki