University of Twente Student Theses

Login

CNN based dense monocular visual SLAM for indoor mapping and autonomous exploration

Steenbeek, Anne (2020) CNN based dense monocular visual SLAM for indoor mapping and autonomous exploration.

[img] PDF
17MB
Abstract:UAVs for indoor mapping applications are normally equipped with bulky and expensive sensors such as LiDAR or depth cameras. However, this task should be performed using light, small and inexpensive platforms, more agile to move in confined spaces. An additional challenge is given by the absence of the GNSS signal that limits the localization capabilities of the UAV. In this research, the real-time indoor mapping capabilities using only a monocular camera installed on a commercial low-cost UAV (DJI Tello) are investigated. The limitations of traditional monocular SLAM approaches are the lack of scale of the scene and the reduced density of points in the generated map. Deep learning methods are nowadays able to estimate depths from single images, although these products are often affected by large outliers. The proposed method integrates SLAM algorithms and CNN-based single image depth estimation algorithms in order to densify and scale the data and deliver a map of the environment, suitable for exploration, in real time. The details of the implemented algorithms, the training strategy of the network as well as the tests on each element of the proposed methodology are reported in this work. The results achieved in a real indoor environment are also presented, demonstrating the potential of this solution in the rapid exploration of unknown environments.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:53 electrotechnology, 54 computer science
Programme:Systems and Control MSc (60359)
Link to this item:https://purl.utwente.nl/essays/81420
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page