Figure 4. Histograms of the residual error of all checkpoints. The
bars indicate the probability of the corresponding residual errors,
with bar intervals of one pixel.
Vertical Accuracy
Study Data
The
ZY-3
satellite can provide two types of stereo images:
three-view stereo images (forward, backward, and nadir im-
ages) and two-view stereo images (forward and backward
images). Since the vertical accuracy was determined by the
stereo intersection angle, the addition of the nadir image
increased the observed value, enhancing the planar position-
ing accuracy. Therefore, compared with the two-view stereo
images, the three-view stereo images did not improve the ver-
tical accuracy (Tang
et al.,
2012; Pan
et al.,
2013). A similar
conclusion was obtained using SPOT5
HRG
(High Resolution
Geometrical) and HRS images to conduct block adjustment
simultaneously (Kornus
et al.,
2006). Therefore, to simplify
the experimental process, the vertical accuracy verification
used only forward and backward stereo images.
Twelve
SC
stereo image pairs from three orbits (four stereo
images from each orbit), covering an area of ~14,000 km
2
near
Taiyuan City, Shanxi Province, China, were used. The valida-
tion and reference data employed the highly accurate digital
elevation model based on aerial Light Detection and Ranging
(lidar) acquisition (
LIDAR
-
DEM
) obtained on 20 February 2013.
The
LIDAR
-
DEM
, covering an area of ~8,000 km
2
, had vertical
and planar accuracies of 0.5 and 1.5 m, respectively, with a
GSD
of 3 m (Figure 5). According to the geometrical distri-
bution from west to east, the acquisition times of the three
orbit images were 30 October 2012, 03 January 2013, and 22
February 2013. Each stereo image pair contained the forward
and backward images and was produced by the postpro-
cessed attitude and orbit measurements. The experimental
area included plain and mountainous terrain, with a variable
elevation of 700 to 2,000 m.
Experiment and Results
With the purpose of obtaining a sufficient number of high-
precision tie points for stereo image pairs, the free network
adjustment of all experimental stereo images based on the
RFM
was conducted with the Pixel Factory software. A total
of 1,532 tie points within the area of the
LIDAR
-
DEM
data were
obtained using an automatic matching from Pixel Factory
software (Bignone, 2003). After the artificial point adjust-
ment, the
RMSE
of the tie point was less than 0.25 pixels, with
a maximum error below 0.6 pixels, during the free network
adjustment process. For the forward and backward experi-
mental stereo images before the free network adjustment,
based on the image coordinate information of the tie points,
the elevation was obtained by forward intersection and the
vertical accuracy was obtained by subtracting the elevation of
Figure 5. Distribution of the experimental images and LIDAR-DEM for vertical accuracy verification. The black frames refer to the distribu-
tion range of the experimental images, the gray areas refer to the distribution range of the LIDAR-DEM, and the black area refer to the
distribution range of the zoomed image
932
December 2015
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
