Browsing by Author "Suomela, Juuso"

Filter results by typing the first few letters
Now showing 1 - 2 of 2
  • Autonomiset pinta-alukset ja niiden käyttökohteet
    (2019-12-20) Suomela, Juuso
     Sähkötekniikan korkeakoulu | Bachelor's thesis
  • Three-dimensional map quality assessment using graph residuals and frequency-based structure extraction
    (2022-10-17) Suomela, Juuso
     Sähkötekniikan korkeakoulu | Master's thesis
    Map building and localizing a robot within it are fundamental problems with autonomous mobile robots. Simultaneous Localization and Mapping (SLAM) approaches aim to solve this problem. State-of-the-art SLAM solutions work in most environments, but noisy data and challenging environments can lead to SLAM failing. As a result of this, the produced map may not match the real environment. Map quality needs to be assessed to decide whether a SLAM-produced map can be used reliably for localization. The aim of this thesis is to develop an approach for assessing the quality of maps based on a 3D Normal Distributions Transform Occupancy Map (NDT-OM) and information obtained from a Graph-based SLAM algorithm. The approach consists of two parts: graph analysis and Robust Frequency-Based Structure Extraction (ROSE). Graph analysis evaluates the topological structure of a map by analyzing pose graphs generated with a Graph-based SLAM. Graph residuals and the optimization score of the graph are used for this topological analysis. ROSE assesses the geometry of a 2D map by filtering it in the frequency domain to clear the noise and identify dominant structures for map quality assessment. ROSE is used for detecting geometrical quality anomalies that graph analysis cannot detect, and the performance of ROSE is assessed in urban outdoor areas. The introduced map quality approach was evaluated using visual evaluation and feature evaluation of NDT-OM with multiple maps from urban outdoor and indoor environments. With these evaluation methods, it was found that graph analysis reliably detects inconsistencies in the graph of Graph-based SLAM. These inconsistencies were further analyzed to observe map quality problems and fix those. ROSE was also successfully used with outdoor maps to detect geometrical map quality issues that graph analysis was not able to detect. A combination of graph analysis and ROSE provides a comprehensive and generic reference-free map quality approach. For future work, the graph analysis could be extended to support more graph formats and automated to provide easily interpretable results.