a, CA, 9-11 Nov. 1999 
on roof detail extraction rather 
The extraction results will be 
structures observed in the field. 
DY AREA 
en that represented alternative 
ial area was chosen because of its 
roofed buildings. If the LIDAR 
[ detail for buildings then it will 
are 1 shows the industrial area as 
with building outlines added for 
tograph of the industrial area is 
> dominant roof structure for this 
It roof split by a central ridge 
»'s long axis. 
to represent a more challenging 
buildings in this area (Figure 3) 
plex roof structures than the 
e is also non-building noise from 
hedges (Figure 4). 
ODOLOGY 
the LIDAR data parameters for 
urvey was undertaken to create a 
f structures. 
These were then 
LIDAR algorithm results using 
sted, an error assessment of the 
| was carried out. This was to 
t into context with any inherent 
  
industrial area buildings. 
    
  
  
Figure 3 LIDAR representation of residential area with 
vector building boundaries (boundaries reproduced from 
Ordnance Survey mapping with the permission of The 
Controller of Her Majesty's Stationery Office, Crown 
Copyright. ED 273554). 
  
Figure 4 Example of residential area buildings. 
À quantitative assessment of LIDAR vertical accuracy was 
made by comparing LIDAR heights against Ordnance Survey 
spot heights from a 1:1250 map. Root Mean Square Error 
(RMSE) was derived from the comparison as a measure of 
vertical error (Jaafar and Priestnall, 1998). The planimetric 
accuracy of the LIDAR data was determined qualitatively using 
the author's own observations as well as available literature. 
The planimetric accuracy of the vector building data was 
extracted from the dataset's metadata. 
3.2 Survey Methodology 
The survey data set is a plan description of the roof structure 
for every building in the LIDAR data set. A visual assessment 
of each building was made in the field, and all roof edges 
including dormers and other small extensions were drawn onto 
à 1:1250 building map. 
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999 
  
   
Vector 
building 
outlines 
   
    
             
        
    
  
LIDAR 
  
elevation, ; : 
slope and Create grid mask of 
aspect building areas 
data 
  
  
  
  
Extract LIDAR 
building pixels 
using mask 
  
  
  
  
       
    
       
Elevation/ 
Slope Aspect 
Which 
parameter 
9 
  
  
  
  
  
  
  
  
Standardise pixel Calculate median 
values for each value for each 
building from building 
0-100956 i 
Y Reclass building 
Select highest pixels 
Sea ae If Value > median, 
(elevation) or lowest 
(slope) x% of values value = 2 
If Value < median, 
t 
  
  
  
  
  
  
  
  
Reclass Selected Extract pixels with 
pixels = 1 value 1 or 2 that 
touch pixels with 
  
values 2 or 1 
  
  
  
  
Convert 
extracted 
pixels to vector 
lines 
Extend vector lines to 
vector building outlines 
  
  
  
  
  
  
  
    
      
      
Vector 
building 
and roof 
detail 
Figure 5 Flowchart of algorithm development process. 
3.3 Algorithm Development 
LIDAR elevation and derived slope and aspect parameters were 
used in the algorithm development. Each parameter was taken 
in turn and manipulated by the algorithms to extract the 
maximum amount of information from it. Figure 5 is the 
standard algorithm used for all three parameters. To save 
processing time the vector building data was used to isolate 
LIDAR building pixels only. Each LIDAR building was then