Figure 7. Example of referenced points location - vector (red) 
and texture (green) elements of model. 
4.4 Absolute accuracy 
For the corresponding control points were calculated 
differences between heights and distances, both horizontal and 
spatial. This allowed to determine vertical, horizontal and three- 
dimensional absolute accuracy of the model. The tachymetric 
measured points have very high accuracy so can be treated as 
error-free. The results of absolute accuracy assessment are 
shown in the Table 2. 
  
  
  
  
  
“Object 1” “Object 2” 
Horizontal 
min. residue 0.004m 0.010m 
max. residue 0.147m 0.362m 
mean residue 0.079m 0.101m 
RMSE 0.086m 0.123m 
Vertical 
min. residue -0.132m -0.331m 
max. residue 0.173m 0.349m 
mean residue 0.090m 0.041m 
RMSE 0.114m 0.134m 
Three-dimensional 
min. residue 0.030m 0.017m 
max. residue 0.216m 0.385m 
mean residue 0.139m 0.159m 
RMSE 0.143m 0.182m 
  
  
  
  
Table 2. Results of absolute accuracy assessment. 
The mean vertical residue for *Object 1" shows that the model 
is placed in the space above 0.09m than the real building. This 
value is about 0.07m smaller than indicated in Table 1. This 
discrepancy results from the fact that the final model was 
created based on the combined point cloud. A large number of 
TLS points available on the scanned surfaces of roofs were 
averaged with ALS data and therefore allowed the building 
model to lower into the real height values. 
The values given in Table 2 show that the absolute horizontal 
accuracy is better than vertical. RMS errors for “Object 2" are 
larger than for “Object 1” — this indicates that model of “Object 
|" is more accurate than models created in “Object 2” area. 
Because there were two types of referenced points measured 
(Fig. 7), therefore there were also evaluated accuracies for 
vector end textures elements of the models. Results of this 
investigation are shown in Table 3. Three-dimensional RMS 
errors for both types of the final models elements are similar 
what proves that accuracy of vector elements was not much 
different from accuracy of textures elements of the final model. 
    
  
  
  
  
  
  
  
“Object 1” “Object 2” 
Horizontal 
RMSE of vectors elements 0.095m 0.127m 
RMSE of textures elements 0.063m 0.116m 
Vertical 
RMSE of vectors elements 0.114m 0.117m 
RMSE of textures elements 0.114m 0.160m 
Three-dimensional 
RMSE of vector elements 0.149m 0.172m 
RMSE of textures elements 0.134m 0.197m 
  
  
Table 3. RMSE for vector and textures elements of models. 
5. CONCLUSION 
3D modelling of buildings from laser scanning data has become 
in recent years more and more productive task. While 
algorithms for modelling and automation of this process are the 
subject of intensive research, the more independent assessment 
of the accuracy of product which is a 3D model is less reflected 
in the literature. This is perhaps the fact that the users of such 
models have not yet defined the accuracy expectations. 
Expectations come down mostly to provide the level of detail 
modelling and visual quality assessment. 
In this paper the problem of assessing the accuracy of the 3D 
model of historical buildings in several cities of south-west 
Poland, created on the basis of airborne and terrestrial laser 
scanning data. Assessment of the accuracy was performed by 
comparing the model with precise reference data obtained from 
tachymetric measurements of modelled object. As a result of 
this comparison, it was found that the modelling RMSE is about 
0.14m and 0.18m respectively for investigated “Object 1” and 
“Object 2”. It is worth to notice that the error of texture 
elements mapping (such as the corners of windows) is at the 
similar level as the error of vector elements modelling. This 
proves a correct projection and resampling of digital photos. It 
seems the obtained modelling accuracy is sufficient for many 
problems occurring in practice. It is also significant that the 
modelling was performed for the Internet presentation where 
volume data set as small as possible is required. 
In addition to assessing the absolute accuracy of the model, 
there was performed an assessment of the internal accuracy of 
the ALS and TLS data sets. This assessment was performed by 
comparing the 3D models created separately from ALS and TLS 
data. The result of the comparison, noticed that the internal 
accuracy is at about 0.2m. In order to achieve more accurate 
modelling it is necessary to fit (transform) airborne laser 
scanning data set into terrestrial laser scanning data set. 
6. REFERENCES 
Akca, D., Freeman, M., Sargent, L, Gruen, A. 2010. Quality 
assessment of 3D building data. Photogrammetric Record, 
25(132), pp. 339-355. 
Awrangjeb M., Ravanbakhsh, M., Fraser, C. S. 2010. 
Automatic detection of residential buildings using LIDAR data 
and multispectral imagery. ISPRS Journal of Photogrammetry 
& Remote Sensing, 65, pp. 457-467. 
Borkowski A., J6zkéw G., Jarzabek-Rychard M., Tymków P., 
2011. 3D modeling of the historical monuments for the Opole 
bj Pi ey 
od 
ON hit 
rn Dy hm €? 3. dd Cun em 
Uter "s 
p Ppt M) 0 
hm CU. e. ud