In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010,1 APRS, Vol. XXXVIII, Part 7B
589
Therefore, changes at the vegetation level could not be
detected. In order to focus on the urban changes, a vegetation
mask, computed from the IKONOS multispectral data, has
been used. This mask has been generated using the
Normalized Differenced Vegetation Index (NDVI). Figure 5
displays the two DSMs used in this article after removing of
the vegetation area applying the vegetation mask.
3.2 Generation of the difference image.
Following the procedure described in subsection 2.2, after the
co-registration of the two DSMs depicted in Figure 5, we
apply the “robust difference” approach to calculate the
difference images, where noise has been reduced and changes
can be better analyzed. As we can see from Figure 5, in the
LiDAR-DSM, the building edges are sharper than in the
IKONOS-DSM. Therefore, we choose a 7x7 window (w=3 in
equations (1) and (2)) in the noise reduction procedure for
this evaluation.
Both the positive (which highlights the new constructed areas)
and negative difference image (which highlights the
destroyed areas) were generated in this step. Figure 6(a)
displays the positive robust image difference obtained when
applying a 7x7 window in the noise reduction step. The dark
blue colour means no change, and the light blue colour
corresponds to height changes of about 20 meters.
3.3 3D change detection maps
The different steps of the proposed 3D change detection
method applied to the city centre of Munich are displayed in
Figure 6. Figures 6 (a-c) present the positive change results.
The positive difference image is displayed in Figure 6 (a).
The corresponding change mask is shown in figure 6 (b).
Figure 6 (c) depicts the final change detection map, under the
assumption that each building has only one height value,
computed as described in subsection 2.3. As can be seen from
the change detection results, 6 new buildings are detected.
The buildings No.4 and 5 are detected as one building in the
mask map (showed in figure 6(b)), due to errors in the DSM
generation procedure. To separate the two buildings, in the
box-fitting procedure, we extract two seed points in this area
based on image eroding result. And we use the original edges
in the box growing procedure, so the image eroding will not
influence the accuracy of the result.
fitting these buildings to regular rectangular shape. This
suggests them to be false alarms (as shown in Table 1).
Therefore, only one building is remaining in the negative
change map.
The same steps have been adopted to detect the negative
changes (presented in figures 6 (d-f)). According to the
negative change detection results, four building masks are
detected. But three of them are in strange shape, with relative
low height values. Also, the box-fitting procedure fails in
Table 1. Change Detection Result Comparison
Figure 6. Change detection results: (a) Positive difference
map; (b) Positive change mask; (c) positive change map; (d)
Negative difference map; (e) Negative change mask; (f)
Negative change map
4. RESULTS AND DISCUSSION
In order to allow a quantitative evaluation of the effectiveness
of the presented methods, and also to study the influence of
the building shapes on the change extraction procedure, we
compare the mask-based change maps to the box-fitting
based ones. A manually annotated change map has also been
included in the evaluation scheme:
No.
Change Type
Mask-Based
Box-fitting Based
Manual Extraction
Height [m]
Area [m 2 ]
Height [ml
Area
KJ
Height [m]
Area [m 2 ]
1
Positive change
19.58
3247
19.80
3077
22.0
2788
2
Positive change
19.10
491
16.46
576
17.00
465
3
Positive change
18.97
3344
19.10
3748
18.8
3694
4
Positive change
18.78
2553
19.55
18.26
1818
2911
19.2
17.90
1377
2289
5
Positive change
16.97
1093
16.6
912
6
Positive change
18.64
1683
18.69
1409
20.2
945
7
Negative change
-33.72
762
-32.42
800
-36.0
1007
8
Negative change
-6.2
1044
—
-
—
—
—
—
9
Negative change
-6.43
578
—
-
-
—
—
—
10
Negative change
-6.30
565
...
-
-
-
-
-