International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
including standard deviation of photo coordination
measurement, From Table 2 can be seen that when the
observed pointed are divided into two groups, a’priori standard
deviation of photo coordination measurement is defined
separately as lum and 22pm for the Block 1 and 12um and
25um for the Block 2, while when the points are not grouped,
there is only one value used : 11pm for the Block 1 and 121m
for Block 2. This is because the points on trees and non-trees
are needed to given different and the thresholds for detecting
and low regions according to the disparity images. Next, the
trees are extracted by Fuzzy C-Mean in the high and low areas,
using a set texture and color features. We demonstrated in a
number of aerial images that the correct rates of extraction trees
are from 95% to 88%. Finally, we have shown that algorithm
can be used to group automatically the observed points in aerial
triangulation. The comparison results illustrate that the
adjustment accuracy is improved using the observed points
grouped in the forest-covered areas. In the future, with this
extraction algorithm ,we would analyse that forest affect the
Table 2. Comparison results of the adjustments in the two different conditions
Description Block 1 Block 2 Percentage(%)
Grouped Ungrouped Grouped Ungrouped
Pre-Std.dev. of photo No. I 11 No.l 12
coord(u m) No.2 22 11 No.l 25 12 Block! Block 3
Std.dev.of photo
coord (uum) 8.08 17.03 6.90 11.61 53 41
RMS control (cm) 7,4,5 11,13,14 45,51,65 66,76,112 36,69,64 32,33,42
RMS check (cm) 20.18.27 30,40,54 65,70,101 80,91,120 33.55.50 19,23,16
Mean Std.dev.of
Objects points(cm) 15,14,26 21,20,36 45,53,63 50,58 75 29,30,28 10,9 11
Max. Std.dev.of
Objects points(cm) 40,35,70 39,35,69 118,180,172 110,172,175 -2,0,-1 -7,-5,2
Min. Std.dev.of
Objects points (cm) 72411 14,14,23 22.23.26 31,32,34 50,50,52 29,28,24
Mean td.dev.of 2027.17 47,40,29 44,46,34 49,50,40 38,33,41 10,8,15
rere 17,17,9 27.26.16 4.1,4.2,3.3 6.4,8.2,5.4 37,65,44 36,49,39
mgon
Max. Std.dev.of 60,54,33 60,52,41 83,82,78 96,85,79 0,-4,20 7,4,1
ext.orient(cm)/(mgon) 35,34,19 35.33.21 53.3.7.5.1 6.0,3.5.5.2 0,-3,10 8,-2,2
Min. Std.dev.of 18,16,9.2 3331.19 30.3222 33.37.30 46,48,53 14,13,27
ext.orient(cm)(mgon) 10,9,5 19.19.10 2.8,2.5,1.4 3 0 27,18 47,52,50 7.7.22
percentage (%)=(U-G)/U , where U and G expresses the accuracy values ungrouped and grouped respectively.
error and the weights, in order to guarantee the connecting
strength of aerial triangulation network that is very important to
obtain good results, decrease the negative influence of the
points on trees. The effect of the points grouped is clearly seen
in the results : the accuracy of photo coordinates in the
observed points grouped as compared with the points
ungrouped is 53% higher for the Block 1 and 41% higher for
the Block 2. The RMS of control points in the observed points
grouped is lower 36% (X),69% (Y ) and 64%(Z) for the Block
1 and 32%(X),33%(Y) and 42%(Z) for the Block 2, but, the
maximum standard deviation of objects in the observed points
grouped is higher 2%(X)and 1%(Z) for the Block land
7%(X),5%(Y)and 2%(Z) for the Block 2. This is due to the fact
that the results of the adjustment by the points grouped is very
well within two region, but it has problems that the threshold
for detecting error of the points on trees is bigger than non-
trees, a few lower accuracy observed points are permitted to be
involved in the adjustment. To prevent some bad observed
points to participate the final calculation, in practice, the
reasonable rate of the a'priori standard deviation of photo
coordination measurement is chosen, according to the terrain
characteristic of blocks. Note however that the main accuracy
measures of the adjustment by the observed points grouped are
better than of the points ungrouped for the two blocks.
4. Conclusion
In this paper, we have proposed a new algorithm to extract
the objects which are higher than surrounding in color aerial
images. At first, the original images are segmented into high
accuracy of digital elevation model in the forest-covered
regions.
References
T.Dang, O.Jamet, H.Maitre. 1995. Applying Perceptual
Grouping and Surface Models to the Detection and Stereo
Reconstruction of Building in Aerial Imagery. IntArchPhRS,
Vol.30, Part 3/1, pp.165-170.
N.Haala. 1994. Detection of Building by Fusion of Range and
Image Data. IntArchPhRS, Vol 30, Part 3/1.
W.Eckstein, C. Steger, 1996. Fusion of Digital Terrain Model
and Texture for Object Extraction, Proceeding of 2™ Airborne
Remote Sensing Conference , San Francisco, 1-10.
R. Kaczynski, J. Ziobro, 2000. Comparison of Semi-Automatic
and Automatic Digital Aerial Triangulation.
IntArchPhPR, Vol. X XXIII, Part 3.Amsterfam, pp.457-461.
L. Pan. 2001. The Study on Forest Area Recognition from
Color Aerial Image and Its Application for Automatic Aerial
Triangulation, Ph.D. thesis, Wuhan University.