B3, 2012 
imized models 
zenerated by 
the SIRADEL 
following the 
al., 2010). The 
containing the 
analysis of the 
he geometrical 
vith reference. 
parameters are 
feature of the 
ly to compute 
trated more on 
trical accuracy 
ometrical error 
f building roof 
ted) and of an 
between the 
  
ermined from 
13 possible 
9. _ 
0.56 [m] 
m 109 of 133 
yondences) 
[m] 
om 505 of 816 
yondences) 
0.92 [m] 
m 142 of 183 
yondences) 
  
the report file 
also provided. 
on a per-pixel 
ofs, and blue 
ackground but 
| were derived 
ce respectively. 
'€ a plane was 
ıyed in white. 
  
pum = e 
pu mines see 
T _ § i i 
1 A - — 
1 UT I 
M b m p 
| "E ' © | 
PE CN) 4 1 
QUE D e 
Wo. = us m TE 
ut a 
Figure 6. Evaluation of building detection on a per-pixel level 
  
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
Figure 7. DSM difference between reference and result 
The observations from these two images and explanations are 
listed below. 
a. The reconstructed roofs fit reference roofs well. We reach 
this conclusion because of regular yellow rectangles in Fig. 6, 
also because the color of most roofs in Fig. 7 is green (green 
stands for small difference between result DSM and reference 
DSM). 
b. In Fig. 6, there exist some blue blocks. Because some 
buildings can not be represented by simple hip-roof primitive, 
they were not reconstructed. And we missed some small planes, 
they were not reconstructed too. 
c. In Fig. 6, there are some red lines around yellow roofs and 
some small rectangles adjacent to yellow roofs. And in Fig. 7, 
there exist some small blue dots on the roofs. Because the 
proposed method is a primitive-based reconstruction method; 
some buildings don't strictly coincide with primitive, and some 
detailed features of buildings can not be represented by current 
simple primitive. 
4. CONCLUSIONS AND FUTURE WORK 
We proposed a primitive-based 3D building reconstruction 
method which can utilize the complementarities of airborne 
LiDAR data and optical imagery. It has not only the merits as 
other model-based methods, but also two characteristics. The 
proposed method is simple because it only uses the most 
straightforward features, i.e. planes of LiDAR point cloud and 
points of optical imagery. Further more, the proposed method 
can tightly integrate LIDAR point cloud and optical imagery, 
that is to say, all primitives’ parameters are optimized with all 
constraints in one step. 
We applied this primitive-based 3D building reconstruction 
method to an ISPRS Test Project data. The evaluating result 
showed the proposed method is feasible. The reconstructed 3D 
building models fit the outlines of reference roofs well. 
At present, the proposed method has some deficiencies. Firstly, 
current simple hip-roof primitive can not completely represent 
actual building, especially detailed features. Secondly, there 
are many manual works. For example, extraction of 2D corner 
features and 3D plane features, selection of primitives and 
measurement of the initial parameters of these primitives. The 
first deficiency can be partially overcame by using more 
primitives such as cylinder, sphere, and so on, and a complex 
building can be represented by CSG (constructive solid 
geometry) model which can be derived by using bool operation 
on these primitives. The second deficiency is the main 
drawback of the work in this paper. The emphasis of the work 
in this paper is to prove that our method can obtain optimized 
buildings by simultaneously using features from images and 
LiDAR point cloud. So there are many manual works 
especially in features extraction procedure. But it should be 
noticed, because only simple features (corners in images and 
planes in point cloud) are utilized, so it will be easier to extract 
these features in an automated way then those complicated