re is an 
here, the 
IS: 
(1) 
alue of Ar 
ition that 
boked up, 
cked. 
' distance 
of target 
reshold is 
ict when 
re is used 
1ally used 
statistics. 
nensional 
| can be 
, moment 
92): 
(2) 
] moment 
(3) 
variance, 
| moment 
(4) 
bution or 
ments are 
r moment, 
du, 1962) 
ormalized 
Moment invariant 4: 
¢, = (uy, +, + (uy, +) (8) 
Moment invariant 5: 
@; = (uy —3u,, us, + 242) < [450 + M2 Yu, + Ms] (9) 
+ (Buty, — Uo3 (t, 1) x [305, * Y- (u5, ug; y] 
Moment invariant 6: 
$, 7 (uy — tty (115, Tug) = (uy, Yi Y] (10) 
* Au Qu uu, us) 
Moment invariant 7: 
d, Qu, — us tss + 24,2) X [Go 1 y -39, tus] (11) 
+ (35, 7 Mg Yt + 2003) < [300650 + Mo). t s Y 1 
Semi-major axis a: 
  
id Up Fg, [ss tg, y Um 1/2 (12) 
a=[ ] 
Hoo /2 
Semi-major axis b: 
52 [420 Tu - [59749 y +Aur, De 1? (13) 
Um/2 
Based on the above parameters, ratio p of semi-major axis and 
semi-minor axis is: 
p=alb (14) 
Radiometric F of ellipse is: 
F=u,/zab (15) 
As the same with Hu's 7 moment invariants, in actual data 
processing, (» and F should be expressed by normalized central 
moments. 
4. AIRPLANE RECOGNITION METHOD 
In LiDAR data, the direction of airplane is arbitrary, the outline 
and size are usually different, but they are similar in appearance, 
so the method must consider deformation and scale invariance. 
We use prior knowledge and invariant recognition moment to 
recognise airplanes (even hidden targets mainly under the cover 
of canopy). Figure 1 shows the processing procedure as follows: 
1. Filter and classify point cloud data into ground points 
and non-ground points; 
2. Organize non-ground points by KD-tree; 
3. Cluster the organized point cloud data and calculate 
the different targets; 
4. Transfer each object into depth image. Assign the 
minimum and maximum of the targets’ height as 0 and 255, 
so the height range can be extended to 0-255; 
5. Refine depth image; 
6. Remove the incorrect results based on prior 
knowledge such as the targets' length, width, height 
and the ratio of length to width; 
7. Calculate seven invariant moments of airplanes; 
8. Identify target by Euclidean distance between the 
target and template which was built before. 
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 
  
f LiDAR point clouds data of 
i 
Point Cloud Filtering 
Cobtain non-ground points ) 
  
  
  
Isolate Individual Objects by Clustering 
  
¥ 
  
Convert into Depth Image 
  
* 
  
Refine Depth Image 
  
  
+ 
Remove the Incorrect Results 
Based on Prior Knowledge 
  
* 
  
Calculate the Moments Invariant of Targets 
  
* 
  
Identify Targets by Euclidean Distance 
  
  
  
Figure 1. Flowchart of airplane recognition 
5. EXPERIMENTAL RESULTS AND ANALYSIS 
In this section, we experimented on the LiDAR point cloud data 
to confirm the recognition method for airplane targets proposed 
in this paper. The point cloud data to be identified is shown in 
Figure 2 (the red and green points respectively represent ground 
and non-ground points), which contains 23128 points and with 
a density of 2.4 points/m?, and the results of target recognition 
are shown in Figure 3 (the green and red points respectively 
represent airplane target and other target points). 
  
Figure 2. Point cloud data to be identified