In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Voi. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009
Figure 5. Road parts extracted in subset 3 (yellow).
The road parts are then assembled into road subgraphs as is
shown in Fig. 6 for subset 1. There are three subgraphs which
contain different hypotheses; these are resolved using linear
programming, as described in Section 2.3. In Fig. 7, only these
three subgraphs are shown with the edges that are removed
displayed in red. The results show that the optimisation favours
connections between road parts that are similar in colour and
width and maintain a more or less straight continuation.
Figure 6. Road subgraphs (without DSM) for subset 1.
Different colours represent different road subgraphs.
Figure 7. Road subgraph evaluation (without DSM) for
subset 1. Discarded gap edges are displayed in red.
3.2 Results with DSM
The grouping and the road part extraction were repeated using
the DSM as additional information, as described in Section 2.4.
Figures 8, 9 and 10 show the results of the road part extraction
with the DSM for the image subsets 1, 2 and 3, respectively.
Both the completeness and the correctness values (Table 4)
have notably improved compared to the results without the
DSM. The highest improvement in completeness is observed in
subset 3; almost all roads are now covered with road parts for
the greater part of their area. The highest improvement in
correctness is observed in subset 1 where several buildings were
extracted without the DSM.
Completeness
Correctness
Subset 1
73 %
73 %
Subset 2
91 %
65 %
Subset 3
45 %
57%
Total
70%
65 %
Table 4. Evaluation of road part extraction with a DSM.
The subgraph generation and evaluation is conducted in the
same way as before. The subgraphs for subset 1 can be seen in
Fig. 11. Now there is only one subgraph with several branches,
because the use of the DSM prevented some buildings from
being extracted. The result of the evaluation of the remaining
subgraph with branches is shown in Fig. 12.
31