The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
728 
Number of Pixels 
Site Accuracy 
(%) 
Number 
of 
Trainin 
g 
Test 
Trainin 
g 
Test 
Classifie 
d Pixels 
Roads 
3686 
1662 
96.7 
96.1 
25512 
Grass 
960 
499 
98.5 
35.3 
2047 
Water 
3331 
7695 
99.6 
99.8 
35012 
Sand 
2299 
4584 
100.0 
99.9 
12319 
Bare 
Soil 
323 
115 
99.7 
98.3 
1312 
Other 
7564 
5041 
84.1 
59.2 
100702 
Total 
18163 
19596 
92.7 
87.6 
176904 
Table 1. Results of the AVIRIS image classification after 
removing the building pixels 
The next step in the process is to generate the road and the 
shoreline vector layers. The road class is first separated from 
other classes then a thinning process is performed on road 
pixels. A 2D Hough transformation parameter space is filled 
using all road pixels. Peak cells in the parameter space are 
located and all pixels contributing to these cells are used to find 
the road centrelines using a non-linear lest squares estimation 
model. A line joining process is then used to join road 
centrelines that have the same parameters. 
In the last step, the shoreline is extracted from the water class. 
First, the water class is separated from other classes and border 
pixels are defined. These pixels are then converted to a polyline 
using the algorithm presented in Bimal and Kumar (1991). The 
basic idea of the algorithm is to go through all the border points 
and only retain those that that are significant, i.e. those that 
represent vertices. Figure 3 shows the extracted road network 
and the extracted shorelines overlaid on the reference 
Figure 3. Extracted road network and shoreline overlaid on the 
reference orthophoto 
7. EVALUATION AND ANALYSIS 
The completeness and the positional accuracy of the results are 
evaluated using different metrics. The completeness is 
measured using two metrics: the detection rate (DR) and the 
false alarm rate (FR). Table 2 shows the values of the DR and 
FR for each class. The table shows that the detection rate for all 
three features is more than 90%. In addition, the table shows 
that no false-alarm roads or shorelines have been extracted. The 
table shows that the false-alarm rate in the building layer is 
3.2%. 
Detection rate (%) 
False -alarm rate 
(%) 
Roads 
91.3 0 
95.2 0 
93.2 3.2 
Shorelines 
Buildings 
Table 2. Quality metrics used to evaluate the results 
The RMS is used to evaluate the positional accuracy of the 
extracted features. This is performed in different means. For the 
road network, the positions of 20 road intersections are used. 
The RMS for the coordinates of the 20 points is 5.7 meters and 
the maximum error is 9.5 meters. For the shoreline, 30 distinct 
points are manually selected. The RMS for the 30 points is 7.2 
meters, and the maximum positional error is 10.9 meters. The 
RMS for the buildings comers is 2.3 meters, while the 
maximum error is 3.8 meters. 
8. CONCLUSIONS 
This research shows that the combination of multiple and 
independent remote sensing data is essential to solve the 
complexity of the coastal mapping. Due to the nature of the 
LIDAR data, it was employed to extract building polygons 
only. The AVIRIS data led to other coastal features that are not 
distinguished using the height attribute only such as roads and 
shorelines. The classification results are vectorized and used to 
generate road and shoreline vector layers. Results show that the 
average detection rate of the proposed technique is 93%. The 
positional accuracy of the extracted features is data depended. 
The research shows that the RMS of the generated building 
polygons is 2.3 meters, while the RMS of the roads and the 
shorelines is about 6.5 meters. These results show the potential 
of merging optical and laser data to provide reliable and 
accurate geospatial information that can be used to build or 
update coastal GIS database. 
REFERENCES 
Ackerman, F., 1999. Airborne laser scanning - present status 
and future expectations. ISPRS Journal of Photogrammetry & 
Remote Sensing, Vol. 54(2-3), pp. 64-67. 
Bartels, M., Wei, H., and Ferryman, J., 2006. Analysis of 
LIDAR data fused with co-registered bands. Proceedings of 
IEEE International Conference on Advanced Video and Signal- 
Based Surveillance, 
http://www.cvg.reading.ac.uk/projects/LIDAR/index.html, 
(accessed 6 Nov. 2007). 
Biehl, L., and Landgrebe, D.A., 2002. MultiSpec: a tool for 
multispectral-hyperspectral image data analysis. Computers & 
Geosciences, Vol. 28(10), pp. 153-1159.