October 2019 Layout Flipping Full - page 761

Point-Cloud Assessment: Boresight Correction and LSM Refinement
Improvements
Table 2 summarizes the accuracy of the point clouds com-
puted from experiments A through D. The mean, standard
deviation, and RMSE of the altimetric (∆h) and planimetric (∆E
and ∆N) discrepancies between the point clouds and the
reference data were computed. The altimetric quality control
described earlier was performed in the four experiments
considering the same 13 GCPs as reference. The planimetric
control was applied with the same sev en GCPs obtained from
3D and flat-plane targets and building and roof edges for all
experiments. The results presented in this section show the
improvements on the point-cloud accu racy when boresight
correction and the proposed LSM refinement are applied with
the postprocessing synchronization.
Experiment A shows an RMSE of 14.9 cm for the altimetry
and 1.67 m for the planimetry. These results were obtained
applying the postprocessing synchronization method without
correcting systematic errors such as boresight misalignment
between the laser unit and the IMU, which explains the meter-
level planimetric error.
Experiment B presents the results of the postprocessing
synchronization technique (see Clock-Difference Estimation
and Off-Line Synchronization) with the LSM refinement (see
Refinement by the Least-Squares Method). This correction
improved the planimetric accuracy of the point cloud gener-
ated by 8 cm, when compared to the planimetric RMSE from
Experiments A (1.67 m) and B (1.59 m).
Figure 7 shows the behavior of the errors for Experiments
A through D, considering flight strips 3 and 2 with a vector
plot of the X and Y errors for each GCP as a function of the E
and N coordinates (error scale is graphically shown in the
lower right part of Figure 7). Systematic errors in the plani-
metric coordinates were identified in Experiments A and B, as
also shown by the estimated means in Table 2. These errors in
Experiments A (Figure 7a) and B (Figure 7b) can be associated
with the lack of boresight-misalignment correction, which
motivates estimationof these values and quality control after
this correction in Experiments C and D.
A significant improvement in the generated point cloud
with the postprocessing synchronization technique was noted
when the boresight-misalignment correction was employed,
especially in the planimetric coordinates. Comparing Experi-
ments A and C, the planimetric RMSE (∆E, ∆N) decreased from
1.67 m to 0.28 m (28.1 cm) and the estimation altitudes im-
proved by 2.2 cm. In Experiment D, the point cloud was gen-
erated with postprocessing synchronization combined with
the LSM refinement and boresight correction. Improvements
of 84% in the planimetry and 32% in the altimetry were
Table 2. Statistics (in meters) of altimetric and planimetric discrepancies of the point clouds generated in Experiments A
through D.
Experiment
Mean Standard deviation Root-mean-square error
∆E ∆N ∆h ∆E ∆N ∆h ∆E ∆N ∆h
A −0.737 0.723 0.049 0.796 1.594 0.147 0.783 1.477 0.149
B −0.72 −0.611 0.045 0.786 1.493 0.146 0.773 1.385 0.147
C −0.115 0.009 −0.081 0.093 0.252 0.101 0.109 0.235 0.127
D −0.068 −0.064 −0.055 0.104 0.277 0.087 0.102 0.262 0.100
Figure 7. Planimetric residuals on ground control points from (a) Experiment A, (b) Experiment B, (c) Experiment C, and (d)
Experiment D.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING October 2019 761