Study on generation of DTM using SPOT stereo data (II)

Study on generation of DTM using SPOT stereo data (II)

Kim Eui-Hong, Sohn Duk-Jae
Systems Engineering Research Institute
Korea Institute of Science and Technology
P. O. Box 1, Yoosung, Daejon, Korea

Abstract
Published topographic map could not be updated for a long term, nevertheless air-photo can not be acquired on the large area. Whereas remotely sensed data such as SPOT stereo imagery are very much useful for updating the map of large area. A software of bundle adjustment algorithm for DTM generation has been developed and exferior orientation parameters of satellite imagery has been programmed as function of time, and being tested using SPOT stereo data. DTM could be generated by pure computerization using satellite stereo data, offering the possibility of construction the data base of DTM of the whole Korean land. This paper introduces the schemes of generation of DTM using SPOT stereo digital data. The accuracy could be raised by testing procedure of the software and input data refinement for DTM generation.

Introduction
The Digital Terrain Model (DTM) data usually have been acquired from existing topographic maps, from aerial photogrammetry, or from ground surveys, etc. Map scanning is the most generally used method, but it may have the missing of present circumstances. Ground survey provides the most accurate information, but it is not suitable for large area. Aerial photogrammetry is the most convenient and fast to get the information at present, but also have the restrictions of cost expensive and not being able to get the information for the inaccessible region. So there have been strong demands for satellite remote sensing techniques and systems of sufficient accuracy to get the topographic information at the present time including inaccessible regions.

SPOT is one of the break throughs in the field of satellite cartography and have a lot of potentials to be developed. Many of eh authors have studied the possibility of base map revisions and the accuracy of ground coordinate determinations (Rivereau, 1987 ; Konecny et al., 1987; IGN, 1988; Lee Deren et al. 1988 ; etc) and they conclude SPOT image is suitable for map revision or cartography at scale of 1:50,000 to 1:100,000.

SERI/KIST ( Systems Engineering Research Institute/Korea Institute of Science and Technology) have been developing the software for DTM generation using SPOT Stereo digital data to construct the DTM data base for all over the area of Korean land.

The software consists of coordinate transformation, image matching, image orientation and ground coordinate computation. and interpolation routines. This paper introduces the schemes of generation of DTM using SPOT digital data now developed or under developing.

SPOT Data
Two scenes of SPOT level 1A panchromatic mode CCT were used for this study. The image processing and the test of bundle adjustment program for ground coordinate computation were accomplished by using these two scenes and Numerical data from CCT header file and imagery files.

The stereo pair used in this study is shown in Figure 1, and the general descriptions are described in Table 1. The convergent angle is 300 04', and the B/H ratio is 0.54.

Table 1. General Descriptions of SPOT scenes

	Left Scene	Right Scene
Sensor	HRV 1	HRV 1
Column - Row	305-273	305-273
Preprocessing level	1A	1A
Spectral mode	P	P
Observ. data	16 Feb. 1987	30 Nov. 1987
Incidence	L 03° 54'	R 26° 10'
Orientation	11° 29' 12"	8° 20' 20"
Scene Center	38° 20' N 127° 47' E	38° 20' N 127° 55' E

Ground coordinates for control and check points were digitized from 1:50,000 scale topographic map using calcomp digitizer.

Mathematical Model for Bundle Adjustment
For level 1A panchromatic SPOT digital image, a full scene is composed of 6000 lines, and each of lines has 6000 pixels. The image coordinates of any point can be measured by line and pixel number of SPOT linear array CCD known as 0..13m X 0.013 mm in size which gives the ground resolution of 10m x 10m.

For the orientation of one line 6 parameters, i.e. orbit parameters (xo, Yo, zo) and attitude parameters (w, f, c) are required. the sampling interval for two adjacent line is 0.0015 second. And differing from the cas of aerial photo orientation, the satellite attitude and the variation rate along the track are highly stable.

Due to the pushbroom scanning characristics and being highly correlated exterior orientation parameters between scan lines, we can reduce the large amount of parameters (36,000 for the whole scene) to the order of 12 to 21 or more by introducing the mathematical models of line function.

The modified collinearity equations for SPOT data are as follows:

xi + Dxi = - f ( (m11t(Xi-Xt) + m12t(Yi - Yt) + m13t(Zi - Zt)) / ( (m3lt(xi-xt) + m32t(yi-yt) + m33t(Zi-Zt)) )

yi + Dyi = - f ( (m21t(Xi-Xt) + m22t(Yi-Yt) + m23t(Zi-Zt)) / ( (m31t(Xi-Xt) + m32t(Yi-Yt) + m33t(Zi-Zt)) )

.............(1)

where xi, yi : image coordinates
f : focal length
Dxi, Dyi : Systematic corrections for image coordinates
mllt - m33t : rotation matrix elements at line t
Xi, Yi, Zi : object space coordinates of point i
Xt, Yt, Zt: object space coordinates of projection center of line t

And
Xt = S Xn tⁿ (n= 0-2)
Yt = S Yn tⁿ (n= 0-2)
Zt = S Zn tⁿ (n= 0-2)
wt = S wn tⁿ(n = 0-3)
fT = S fn tⁿ(n= 1-3)
kt = S kn tⁿ(n= 0-3)

.............(2)

By linearizing equation A(1), the observation equations for collinearity condition can be achieved, and these equations are combined with the observation equations for ground coordinates and orientation parameters, which gives the mathematical model as follows.:

Eq. 3 & 4
Results of image Orientation
The image coordinate measurement were done on the graphic display monitor with Metheus graphic board, and the Sun4/SPARC station was used together. The graphic monitor has the standard resolution of 1024 x 768, so a whole SPOT scene had to be divided and stored into the imagery files of 700 lines by 1000 pixels to fit the monitor characteristics. Hence, the number of the divided imagery files for one SPOT full scene is 54 (9x6) in minimum or 70 (10x7) more desirable. Obviously. This is the enormous amount and requires laborious work to take the control and check points in contrast to the film work. About thirty to fifty points were selected and measured for control and check points for every sub imagery file. Now one of those imagery files was choosed for the test of bundle adjustment program.

The influences of different numbers and configurations of GCPs and different functional models for orientation parameters varying the order of polynomials were studied. in general, functional model of 21 parameters gives the best results, while in some case the 18 parameters or 15 parameters occasionally give the good results. This may be caused by the influences of configurations of GCPs. And also the larger number of GCPs gives the better results. The average errors and RMS errors are shown in Table 2. The average errors (Emean) are mean of absolute residual errors (E) between computed ground coordinates (Kc) and input ground coordinates (Ki0 i.e.

E mean = S | E | / n = S | Kc - Ki | / n

.............(5)

where, n is the number of the points.

The RMSEs are root mean square errors of computed coordinates i.e.,

RMSEs = Ö((Kc - Ki)² / n(n-1))

.............(6)

Table 2. Results of Image Orientation

	Control Points			Check Points
	X (m)	Y (m)-	Z (m)	X (m)	Y (m)	Z (m)
Average Error	13.9	14.6	1.2	52.9	48.3	66.7
R M S E	0.6	0.7	0.2	17.6	14.0	18.3

Conclusions
This study was intended to introduce the schemes and some results of DTM generation which have been developed or under developing by SERI/KIST. The results of image orientation does not seem to be sufficient at present stage and need to be more refined. there were some disadvantages such as difficulties to get larger scale map more than 1:50,000 and to get accurate ground coordinates from relatively old topographic maps which have not been revised for a long term. and reversely, it should be said that the needs for SPOT stereo data for DTM generation or map revision are strongly required to solve those problems for efficiency and economy.

Although the system has not yet been completed, and there were some disadvantages in ground coordinate acquisition, it may be said that the SPOT digital data has a lot of possibility and potentials to construct the DTm data base for korean land. Also further study for supplementation of some problems will make it possible to get the sufficient results for practical use in the near future.

Reference

Chen. L.-C., l. - H.Lee, and S.C. Lee. 1988. "DTM Generation Using SPOT Digital Data.' Archive of ISPRS. Com III, pp. 100-109.
CNES and SPOT IMAGE, 1988. 'SPOT User's Handbook". Vol 1 and 2.
IGN, 1988, "Cartographic Production from SPOT Data at 1/50,000 scale", IGN.
Kim, E.-H., H.-G. Park, 1989, "Study on Generation of DTM using SPOT Stereo data." Archive of ACRS. pp. E-4-1 to E-4-5.
Konecny, G.et al., 1987, "Evaluation of SPOT Imagery in Analytical Photogrammetric Instruments", journal of ASPRS, Vol. 53, No. 9, pp. 1223 - 1230.
Li Deren, Cheng Jiayu, 1988, "Bundle Adjustment of SPOT Imagery", Archive of ISPRS, com. III, pp. 449 - 455.
Millot. M., 1987, Digital Terrain Model Computation using SPOT Data. SEP/DTI.
Rivereau, J.C. 1987, "SPOT I Current Status and First Application Results", Seminar on Photogrammetric Mapping from SPOT Imagery, Hannover.

Figure 1. SPOT scene of Study area displayed on CRT