) 
0 cm ranges; 
extracted (b). 
uted on a point 
ults, shown in 
y on the range. 
is lower than 
r Flash LiDAR 
1 m and 2.3 m, 
jbers are small, 
Note that the 
tion method, as 
  
(b) 
(a) 1 m and (b) 
plane residual 
xameters were 
h range. Figure 
r envelope and 
b). The results 
at the shortest 
object distance, the STD is lower than 1 mm and the maximum 
error is 1 cm, while at 3.5 m (the ambiguity limit) the STD is 7 
mm and the maximum is 5 cm. Theoretically, the STD function 
should be of quadratic form based on the used calculation 
method, yet the curve looks almost linear. Normalized for the 
range, the STD is about 0.2% of distance while the maximum 
error is about 1.6%, as shown in Figure 6b. More details of this 
test are explained in (Toth et al., 2012). 
Accuracy function {STO 
  
  
  
£ We — vm Mea En AYS — axe dn 
s Dsstance « SECUREY vétio 
E M A A 
i / ; 
dd. AMAAZVNN. 
A d \ 
(b) 
Figure 6. STD of residual surface fitting errors (a), and 
normalized statistical parameters (b). 
32 Sphere Fitting Test 
Based on the good experiences with plane fitting, a second test 
was performed. Instead of plane fitting, two spheres with a 
radius of 30 cm were measured. The test range was from 0.7 to 
4 m with a step of 10 cm and each measurement was repeated 
10 times. This type of measurement yields a better relative 
accuracy characterization and additional information can be 
collected about the effect of incidence angle. The spheres were 
directly connected to each other (Figure 7). 
Filtered Depth image - 100cm 
  
  
190 ao 200 4% san ex 
[px] 
Figure 7. Filtered depth image 
     
   
  
  
  
  
  
  
  
    
    
   
    
  
    
   
   
   
    
   
    
   
    
   
  
    
  
   
   
   
  
  
  
  
  
     
Fitted spheres - 100cm 
206 -: 
  
à 
7734200 
Sn a 5960 
3m " EL 
Figure 8. Fitted spheres (red points with residuals over 5 mm). 
During the sphere fitting, the radius, center point and the fitting 
residuals were calculated. Obviously, the estimated radius 
should be comparable with the directly measured one, and 
similarly, the center point distances are also computable and 
should be twice the radius (Figure 9). 
Radius 
Farms proni 
  
e — 21 i 
Repeatabotyaene | 
ha ten 152 E The 205 25 406 
distance iere] 
(a) 
Sphere distances 
  
T 
£ 
= su X 
5 i 
: \ 
s E 3 
E aA 
© 
2 i } s 
V 
260 
#4 M 38 = E À Ax 
  
d ES 
dita fom 
(b) 
Figure 9. Radius of fitted spheres (a), and sum of radiuses and 
center point distances (b). 
If a camera calibration (Khoshelham, 2011) is performed for 
both cameras (depth and RGB), the accuracy can be increased. 
Figure 9 shows an interesting result: the further objects the 
more down scaled. A scaling factor can be determined based on 
the object distance. The device internal calibration data should 
be extended with precise camera calibration and a scaling factor 
as function of distance should be introduced. The center point 
distance lacks this scaling error and shows somewhat better 
results. Figure 10 shows that the incidence angle has no or little