Izumi KAMIYA Geometric Characteristics of the Early

processing software. These 117 known ground points in total are used for control points or verification points. Figure 3 shows a sample of the record ...

0 downloads 252 Views 745KB Size
Geometric Characteristics of the Early Products of ALOS PRISM

75

Geometric Characteristics of the Early Products of ALOS PRISM Izumi KAMIYA Abstract An orientation program developed for ALOS PRISM was applied to a triplet of the early product of ALOS PRISM to clarify the geometric characteristics of the sensor. Errors of single image observation were 1.0 km before adjustment. The geometric errors were likely caused by rotation of radiometers and by mis-alignment of CCDs on the focal plane. Adjusting the rotation of the radiometers, residuals of single image observation were 4.9 m in the horizontal, and residuals of triplet image observation were 2.9 m in the horizontal and 3.2 m in the vertical, respectively. East–west distributed control points are necessary to adjust the radiometer rotation, especially two pairs of east and west control points are recommended. 1. Introduction

worked well for the simulation data, and clarified error

JAXA (Japan Aerospace Exploration Agency)

factors of the simulation data (Kamiya, 2005, 2006).

launched the ALOS (Advanced Land Observing

The orientation program was applied to early

Satellite) on January 24, 2006. The satellite has three

products of PRISM, which was obtained and produced in

earth observation sensors: PRISM, AVNIR-2 and

the calibration/validation phase. This paper reports the

PALSAR.

result of the application.

PRISM Instruments

(Panchromatic for

Stereo

Remote

Mapping)

Sensing

consists

A DEM/orthoimage generation program was also

of

developed and checked successfully using simulation

forward-looking, nadir-looking and backward-looking

data from ADS40 (Kamiya, 2006). The orientation

radiometers. PRISM observes the ground from 3

program and DEM/orthoimage generation program are

directions within an orbit using the 3 radiometers. Each

expected to enable mapping from PRISM images

radiometer has 6 or 8 CCDs on its focal plane (Earth

without requiring a digital stereo plotter.

Observation Research Center, JAXA, 2006). Usually, 4 CCDs are used for a radiometer. Pixel size is designed to

2. Algorithm

be 2.5 m. One of the most important objectives of

The adjustment of the orientation is a kind of

PRISM is medium-scale mapping and DEM production

bundle adjustment for a push-broom sensor. Though the

without ground control points.

position and attitude of the satellite are provided in the

The position of the satellite is obtained by GPS

standard product of PRISM and these data are expected

receivers, and the attitude by star trackers and gyros. In

to be accurate, the adjustment may assume errors of the

addition, ADS (Angular Displacement Sensor) is directly

position and attitude as polynomials of time.

mounted

on

PRISM

to

measure

high-frequency

oscillation.

The adjustment always assumes errors of image observation, and may additionally assume the following

An orientation program with open algorithms for

errors: position and attitude of the satellite, ground

ALOS PRISM was developed in order to determine error

coordinates of the control points, mounting angles of the

factors and to improve geometric accuracy. The program

radiometers, the principal distances, and the principal

was verified before the launch using simulation data

positions.

which were obtained from LHSystems' ADS40 airborne

Corrections of aberration, atmospheric effect, and

digital sensor (Eckardt et al., 2000). ADS40 is a

earth rotation are necessary for absolute orientation of

three-line stereo sensor like PRISM. The program

PRISM images. These effects are, however, almost the

76

Bulletin of the Geographical Survey Institute, Vol.54 March, 2007

same in a radiometer, which means no effect after the

3.2 Data extraction from level 1B1 product

adjustment using ground control points. The largest one,

Orbit and attitude data must be extracted from the

effect of aberration, was evaluated to be a maximum of

level 1B1 products to orientate PRISM images referring

20 m. Therefore, these corrections are not actually

to the format specification (Earth Observation Research

implemented now.

Center, JAXA, 2006). Orbit data are recorded as positions and velocities

3. Used data

of every 60 sec. Hermite interpolation, which is

3.1 PRISM data

polynomial interpolation satisfying the position and

A triplet set of PRISM data observed at Fukuoka,

velocity of sample points, was executed using 4 sample

Japan (Fig. 1) was used. The data are standard products

points around the scene center. Because time to the

in level 1B1, which is radiometrically corrected and

power of 7 must be calculated in the Hermite

geometrically uncorrected. The data are some of the

interpolation, relative time to scene center in minutes

early PRISM products which have "ALOS precision

was used to reduce the calculation error.

attitude determination value." ADS data were, however, not used to determine the attitude.

Attitude data are recorded as quaternion in ECI (Earth Centered Inertial Coordinate System). Care is required, because the definition of quaternion in the ALOS product differs from usual. Quaternions were converted into rotation matrixes, interpolated in time space, and converted into ECR (Earth Centered Rotating Coordinate System). Information to convert from ECR to ECI is also supplied by the level 1B1 product. 3.3 Known ground points Known ground points selected and measured by staff of the Topographic Development Office, GSI (Geographical Survey Institute) were used. They selected 49 locations within the triplet image, then selected 2 or 3 well-recognized ground points at each location. Ground coordinates of the points were measured by RTK-GPS (Fig. 2). Image coordinates of the points were measured by image interpretation on general-purpose image processing software. These 117 known ground points in total are used for control points or verification points. Figure 3 shows a sample of the record of known ground points; the location includes 3 known ground points.

Fig. 1 Used PRISM data and its location

Geometric Characteristics of the Early Products of ALOS PRISM

Fig. 2 Observation of ground coordinates

4. Results 4.1 Before adjustment Before the adjustment, the residuals of image observation were as listed in Table 1. The values are converted into corresponding ground length in this paper. Residuals of the verification points after the intersection, coordinates of the bundle intersection point minus measured ground coordinates, are listed in Table 2, and their residual vectors are shown in Fig. 4. All known points were used as verification points. Residual vectors of Fig. 4 are almost the same for each radiometer, suggesting shift of the images. 4.2 Shift of the principal positions An infinitesimal shift of the principal positions, an intersection of the optical axis and the focal plane, acts as a horizontal shift in ground space. I adjusted the principal positions of the 3 radiometers and the image observations using all known ground points as control points. Residuals

of

image

observation

for

the

adjustment are listed in Table 1. Residuals of the control points after the intersection are listed in Table 2, and their residual vectors are shown in Fig. 5. Residual vectors of Fig. 5 appear to be whirled. This suggests rotation of the radiometers on the optical axis, which cannot be derived from the shift of the principal positions.

Fig. 3 Record of known ground points

77

Bulletin of the Geographical Survey Institute, Vol.54 March, 2007

78

Table 1 Residuals of image observation

Before the adjustment

Adjustment of

Adjustment of rotation

principal positions

of the radiometers

E (Easting)

677.5 m

N (Northing)

749.5

14.6

2.6

E2

1010.3

15.3

4.9

R

R2

4.5 m

4.1 m

Note: Values are converted into corresponding ground length. The residuals are for verification points for "before the adjustment," otherwise for control points. Table 2 Residuals of verification/control points after the intersection

Before the adjustment

Adjustment of

Adjustment of rotation

principal positions

of the radiometers

E (Easting)

120.2 m

2.0 m

2.1 m

N (Northing)

713.1

8.2

2.1

H (Height)

258.5

16.1

3.2

723.2

8.4

2.9

R

E2

R2

Note: The residuals are for verification points for "before the adjustment," otherwise for control points.

Fig. 4 Residuals of image observation (before the adjustment)

Geometric Characteristics of the Early Products of ALOS PRISM

Fig. 5 Residuals of image observation (adjustment of principal positions)

79

Fig. 7 Residuals of image observation (adjustment of rotation of the radiometers)

Bulletin of the Geographical Survey Institute, Vol.54 March, 2007

80

4.3 Rotation of the radiometers

I adjusted the rotation of the radiometers and

An infinitesimal rotation of the radiometers in satellite coordinates space acts as a horizontal shift and

the image observations using all known ground points as control points.

horizontal shear deformation in ground space (Fig. 6).

Residuals of image observation for the

Because PRISM is a push-broom sensor, yawing, which

adjustment are listed in Table 1. Residuals of the control

is rotation on the satellite Z axis, causes shear

points after the intersection are listed in Table 2, and

deformation unlike a frame sensor.

their residual vectors are shown in Fig. 7.

Fig. 6 Effect of radiometer rotation on ground space

5. Discussion

sum of squared average and squared SD. The RMSEs are

5.1 Residual after adjustment of radiometer rotation

considered as absolute errors and are listed in Table 3.

Residual

vectors

of

Fig.

7,

especially

backward-looking ones, seem to depend on pixel number. The boundaries between the CCDs are drawn on backward-looking of Fig. 7.

The SDs are considered as relative errors and are listed in Table 4. The absolute errors correspond to residuals of image observation before the adjustment, as listed in the

Error vectors of CCD 1 tend to be upward; those

second column of Table 1. The relative errors correspond

of CCD 2 tend to be neutral at left and leftward at right;

to residuals of image observation after adjusting the

those of CCD 3 tend to be leftward at left and rightward

rotation of the radiometers, as listed in the last column of

at right; and those of CCD 4 tend to be rightward.

Table 1.

These tendencies can be explained as shift or

Relative errors were slightly better than those of

linear error of CCD alignment on the focal plane. The

Tadono et al. (2006), but absolute errors were much

positions of 2 points near the ends of the CCDs were

worse. The difference might have been caused by

measured to determine CCD alignment before the launch.

improvement of JAXA's processing system, especially

Both the error of the pre-launch measurement and the

geometric parameter after processing of the Fukuoka data,

deformation after the measurement may cause the shift

or by wrong implementation of my program.

error and the linear error.

The onboard clock of ALOS had a 1-second error causing 7–8 km along track error, which was

5.2 Comparison with other research

repaired on September 22, after the observation of

Tadono et al. (2006) reported geometric errors of

Fukuoka (Tadono et al., 2006). The clock error was

PRISM images. They reported errors in the form of

corrected during the ground processing for Fukuoka's

average and SD (standard deviation). I calculated

data. However, the clock error caused wrong attitude

RMSEs (Root Mean Square Errors) by square-rooting the

control (Tadono et al., 2006), which might affect attitude

Geometric Characteristics of the Early Products of ALOS PRISM

81

determination and cause the difference of absolute errors.

Tadono et al. (2006) also reported errors of CCD

The effect of ignoring aberration, atmospheric

alignment and plotted error values of the backward

effect, and earth rotation is too small to explain the

radiometer against pixel number. The result is similar to

difference.

Fig. 7. Table 3 Absolute geometric error of PRISM images reported by Tadono et al. (2006)

Forward-looking

Nadir-looking

Backward-looking

X (Along track)

10.9 m

18.7 m

Y (Cross track)

63.2

30.4

7.4

64.1

35.7

29.2

R

E

2

R

2

28.2 m

Table 4 Relative geometric error of PRISM images reported by Tadono et al. (2006)

Forward-looking

Nadir-looking

Backward-looking

X (Along track)

2.3 m

1.8 m

2.2 m

Y (Cross track)

5.9

5.4

4.7

E2

6.3

5.7

5.2

R

R2

5.3 Necessary number of GCP

6. Conclusions

Many control points were used in section 4 to

The early product of ALOS PRISM had errors

clarify the geometric characteristics of PRISM. However,

likely caused by rotation of radiometers and errors likely

we cannot use so many control points in actual works.

caused by mis-alignment of CCDs on the focal plane.

The number of control points needed for geometric

Adjusting the rotation of the radiometers, residuals of

correction of PRISM images by rotating radiometers is

image observation, which correspond to the accuracy of

considered below.

single image observation, were 4.9 meters. Residuals of

The effects of the radiometer rotation contain

control points after the intersection, which correspond to

shear deformation as shown in Fig. 6. Control points

the accuracy of triplet observation, were 2.9 m in the

must be distributed on both the left and right parts of the

horizontal and 3.2 m in the vertical.

image to detect shear deformation, that is, an east–west distribution of control points is necessary. The number of unknown parameters is 3 for each

East–west distributed control points are necessary to adjust the radiometer rotation. Two pairs of east–west distributed control points are recommended.

radiometer, and one control point observes 2 values, pixel number and line number, for a radiometer.

Acknowledgment

Therefore, at least 2 control points are necessary to

I thank the Topographic Development Office,

determine the parameters. There is, however, only one

GSI for providing the known ground points data of

redundant observation for a radiometer, which is very

Fukuoka, which were essential for this study. I also thank

dangerous. Consequently, it is recommended to use 2

JAXA for providing the PRISM data and related

pairs of east–west distributed control points, i.e. 4 points

technical information under the collaboration agreement

in total, which have 5 redundant observations for a

between GSI and JAXA.

radiometer.

Bulletin of the Geographical Survey Institute, Vol.54 March, 2007

82

References

Orientation of ALOS PRISM Images and its

Eckardt, A., Braunecker, B., and Sandau, R. (2000):

Verification Using Simulation Data, Journal of

Performance of the imaging system in the LH systems

Applied Survey Technology, No. 16, pp. 76–86 (in

ADS40 airborne digital sensor, International Archives

Japanese).

of Photogrammetry and Remote Sensing, Vol. XXXIII, Part B1, pp. 104–109. Earth Observation Research Center, JAXA (2006):

Kamiya, I. (2006): Development of Orientation and DEM/Orthoimage Generation Program for ALOS PRISM, ACRS 2006 CD-ROM Proceedings.

ALOS/PRISM Level 1 Product Format Description,

Tadono, T., Shimada, M., Hashimoto, T., Murakami, H.,

Revision J, pp. 2-5–2-6, Appendix 2-7–2-19, and

Takaku, J., and Mukaida, A. (2006): Results of Initial

Appendix 3-1–3-2,

Calibration and Validation for ALOS Optical Sensors

http://stage.tksc.jaxa.jp/eorcalos/PRISM_L1_J_ENa.

(PRISM and AVNIR-2), Proceedings of the 41st

zip (accessed on December 19, 2006).

Conference of the Remote Sensing Society of Japan,

Kamiya, I. (2005): Development of a Program for

pp. 129–130 (in Japanese).