September 2019 Full - page 640

the major surfaces. The average bias of the 3D point clouds
is 9.3 cm (see Table 2). The poor results concentrated on the
right part of the ground points because in that location, the
terrestrial laser scanner was installed at ground level, so the
point density and the quality of the laser points on the ground
were poor because of the large incidence angles of the laser
ray (Ge 2016).
Table 2. Assessment of the subway station dataset.
Metric
Bias (cm)
Max. Min. Ave. RMSE
Geometric accuracy of structural edges 22.8 2.0 8.4 4.8
Comparison of point clouds 28.9 0.0 9.3 11.2
Conclusions and Discussion
In this paper, we presented a novel solution, integrating a
SLAM point cloud model, and a SfM + MVS point cloud model
(i.e., image-based model) to generate enhanced and extended
3D mapping results using a low-cost RGB-D sensor. In the
Figure 17. 3D comparison of the subway station dataset between the obtained 3D point clouds and the ground truth point
clouds collected with a Leica BLK 360 scanner.
Figure 15. Biases of the enhanced3D mapping results compared with the geometric groundtruth in terms of the footprint.
Markings have the same meaningsas those in the corridor dataset.
Figure 16. Object’s details from the merged enhanced 3D
mapping results.
640 September 2019 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING