In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
In: Wagner 1 
124 
As shown in Fig 10, the detection results show the main- 
structure changes. The blue models show the correct detection 
results, the red models show the omission results. As shown in 
Fig 11, the detection results indicate micro-structure changes. 
The blue models show the correct detection results, the red 
models show the omission result. The detection results with no 
change are shown in Fig 12. The blue models show the correct 
detection results, the red models show the omission results. 
Threshold 
value 
Length ratio 
0.7 
Angle 
15° 
Distance between lines 
10 pixels 
Table 9. Threshold for line comparison 
X 10 s 
Fig 10. Detection result showing main-structure changes 
Fig 12. Detection result of no changed 
TEST 
Ground 
Truth 
Main- 
structure 
changed 
Micro 
structure 
changed 
Unchanged 
Total 
Main-structure 
changed 
25 
0 
10 
35 
Micro-structure 
changed 
0 
24 
1 
25 
Unchanged 
13 
10 
409 
432 
Total 
38 
34 
420 
492 
Diagonal total 
458 
Overall 
0.931 
Producer's 
0.714 
0.960 
0.947 
0.874 
User’s 
0.658 
0.706 
0.974 
0.779 
Kappa 
0.714 
Table 13. Error matrix of detection with single-threshold 
strategy 
x 10 s 
Fig 11. Detection results of micro-structure changes 
TEST 
Ground 
Truth 
Main- 
structure 
changed 
Micro 
structure 
changed 
Unchanged 
Total 
Main-structure 
changed 
31 
0 
4 
35 
Micro-structure 
changed 
0 
24 
1 
25 
Unchanged 
8 
7 
417 
432 
Total 
39 
31 
422 
492 
Diagonal total 
472 
Overall 
0.959 
Producer's 
0.886 
0.960 
0.965 
0.937 
User's 
0.795 
0.774 
0.988 
0.852 
Kappa 
0.829 
Table 14. Error matrix for detection with double-threshold 
strategy 
After this, v 
test site, 
performance 
i.e., double- 
the single tl 
of the detec 
the user’s £ 
threshold st 
the detectio 
user’s accui 
the double- 
To scrutini 
représentât! 
the detectic 
look at two 
and buildin 
Aerial 
Image 
Lidar 
points 
in the 
building 
model 
Building 
model 
We obser\ 
two buildi 
classified 
case (a), it 
micro-stru 
points on 
incorrect c 
LIDAR p< 
are the pc 
points are 
points are