Drift Correction for SLAM Point Clouds: Smooth Natural Neighbor Interpolation
Kontinen, Kaapo (2023)
Kontinen, Kaapo
2023
Tieto- ja sähkötekniikan kandidaattiohjelma - Bachelor's Programme in Computing and Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2023-05-15
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202305145738
https://urn.fi/URN:NBN:fi:tuni-202305145738
Tiivistelmä
SLAM methods generate three-dimensional models of real-world environments. These models are used in self-navigating robot development. As an example, an indoor space at Tampere University was scanned using SLAM, but the resulting model is distorted. Distortion may be detected by comparing with a publicly available floorplan image. We explain the model's features and the tool used to create it. Based on our findings, we propose an algorithm for post-processing architectural models to fit floorplans.
Three types of error are identified: single-scan, loop closure, and alignment of successive point clouds. Small misalignments cause drift over time, as future sensor positions are computed from past ones. This bends the point cloud model. Drift errors are difficult to correct because humans can spot only them after a model has reached a certain distance.
Our post-processing algorithm is based on Delaunay triangulation. Markers are manually placed to represent current and desired positions respectively. Each marker has a location, rotation, and scale. The algorithm is implemented in the open-source mesh editor Blender. With 13 pairs of markers, a consumer-grade laptop processes 17 million points in 40 minutes with a naive non-parallelized approach. The largest distance between corresponding points in model and floorplan decreases from 3.5 meters to 0.35 meters. Because the algorithm only changes locations of points, it can modify any mesh model as well.
Three types of error are identified: single-scan, loop closure, and alignment of successive point clouds. Small misalignments cause drift over time, as future sensor positions are computed from past ones. This bends the point cloud model. Drift errors are difficult to correct because humans can spot only them after a model has reached a certain distance.
Our post-processing algorithm is based on Delaunay triangulation. Markers are manually placed to represent current and desired positions respectively. Each marker has a location, rotation, and scale. The algorithm is implemented in the open-source mesh editor Blender. With 13 pairs of markers, a consumer-grade laptop processes 17 million points in 40 minutes with a naive non-parallelized approach. The largest distance between corresponding points in model and floorplan decreases from 3.5 meters to 0.35 meters. Because the algorithm only changes locations of points, it can modify any mesh model as well.
Kokoelmat
- Kandidaatintutkielmat [7052]