Robotic mapping system for freshwater and forest environments
Loading...
URL
Journal Title
Journal ISSN
Volume Title
Sähkötekniikan korkeakoulu |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Authors
Date
Department
Major/Subject
Mcode
ELEC3025
Language
en
Pages
94
Series
Abstract
Modeling river and forest environments as point clouds can enhance our understanding and management of them. Mobile mapping systems enable efficient data collection, but errors in the localization of the systems over time adversely impact the accuracy and consistency of collected models. This thesis proposes a rotating platform to enhance a survey-grade two-dimensional laser scanner for mobile mapping. Using a post-processed Global Navigation Satellite System and Inertial Navigation System (GNSS/INS) solution as a starting point, two Simultaneous Localization and Mapping (SLAM) methods are developed, to improve the consistency as well as accuracy of the system. A factor graph was used to combine the GNSS/INS solution with scan registration results using normal distributions transform point to distribution algorithm. Two separate experiments were conducted to take measurements in a river and forest environment for evaluating the proposed system and method. The point cloud models are evaluated qualitatively as well as quantitatively using Rényi Quadratic Entropy (RQE) as a measure of their consistency. Additionally, a ground truth trajectory was collected using a total station to evaluate the absolute trajectory error (ATE) of the GNSS/INS and SLAM localization solutions. The developed system functioned successfully to measure both the forest and river environments capturing trees of up to 30 m. Significant object duplication was found in the point clouds created with the GNSS/INS solution. The GNSS/INS solution had an ATE of 0.1055 m and a maximum error of 0.375 m. A bias in the orientation was discovered and was deemed as the primary cause of errors in the generated models. The developed SLAM method was able to remove the bias when sufficient loop closures were added, consequently improving the entropy by 13.7 percent and 23.4 percent for the forest and river data, respectively. Additionally, the developed SLAM method slightly improved the absolute localization accuracy by 0.6 percent and decreased the maximum error by 7.2 percent. As a result, the consistency of the generated models improved significantly.Description
Supervisor
Kyrki, VilleThesis advisor
Hyyti, HeikkiKaartinen, Harri