International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B4, 2012 Inte 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
    
  
    
  
  
  
  
  
  
Test | Reference | Number Terrain | Off-terrain | Width | Length Density Terrain Features 
Site Sample of Points Points Points (m) (m) (points/m?) 
City Sampll 38010 21786 16224 133.89 | 302.73 0.94 A mixture of vegetation and 
Site 1 | Sampl2 | 52119 | 26691 25428 | 20438 | 264.22 0.97 dia on step hillside, dato 
Samp21 12960 10085 2875 123.79 | 115.19 0.91 Large buildings, irregularly 
City Samp22 32706 22504 10202 187.87 | 181.23 0.96 shaped buildings, road with 
Site 2 Samp23 25095 13223 11872 146.18 | 205.9 0.83 bridge and small tunnel, data 
Samp24 7492 5434 2058 121.86 | 72.44 0.85 gap : 
Ciy: |-:sampir/| 288627 | 15556 13306 | 174.17 | 161.94 1.02 Densely packed buildings win A. 
Site3 vegetation, data gaps. |. 
City Samp41 11231 5602 5629 167.19 | 104.71 0.64 Railway station with trains (low i 
Site4 | Samp42 | 42470 | 12443 30027 | 227.12 | 202.98 0.92 eu of temrain points), data 2 
Samp5l 17845 13950 3895 232.4] | 429.87 0.18 Steep slopes with vegetation, Ik 
Forest Samp52 22474 20112 2362 450.01 | 301.12 0.17 quarry, vegetation on river 2. 
Site5 | Samp53 34378 32989 1389 430.42 | 472.93 0.17 bank, data gaps 2k 
Samp54 8608 3983 4625 185.84 | 267.49 0.17 3 
Forest | smapgi | 35060 | 33854 36. .|59425.| 4592... 04g,» [ LTE Buildings, roads wi: sk 
Site6 embankments, data gaps. 
Forest |. Samp71 15645 13875 1770 394.83 | 221.12 0.18 Bridge, underpass, roads with I 
Site7 embankments, data gaps. 7] 
Table 1. Features of the ISPRS dataset (Sithole and Vosselman, 2003) 
Table : 
Quantitative Assessment Since the 
100% comparison 
9 lowest erro 
90% d 2 which mea; 
S094 irre : ; t e better than 
5 : | |] : enough to c 
70% - à | EDT : = Vosselman, 
60% + # Type I error Group 2 is 
50% - shows a ver 
40% : 8 Type II error errors. Grot 
o - eee " percentage 
3094 4. » À — Accuracy Rate probably ha 
20% + 8 it kappa [s 
10% +- i 100% pu 
Na SV En sou Yr ev dum doc NN S 609, i— 
NS NM ur e im o vo ceo 5 9 | 
8 s $ SG WS SA & 
S = am e = S P FX S e e S & e p 40% + 
20% Im 
0% im 
Figure 3. Type I errors, Type II errors, Accuracy Rates and Kappa coefficients of the 15 sample sites from ISPRS tested by the S 
proposed MTF method INC 
SS «3 
Quantitative Analysis parameters. For example, “samp1 1k” in Table 2 refers to the 
parameter combination for the Kappa coefficient value of 
The 15 sample sites from ISPRS WG III/3 are selected on city Sample 11 showed in Figure 3, the numbers of Type I, Type II Fi 
and forest areas, the ground features such as slope gradient, errors and accuracy for Sample 11 are generated by the igure 4. A 
vegetation density are various. Therefore, Meng et al. (2010) parameters listed as “samp1 1”. 
divided the 15 ISPRS study sites into three groups. The sites in Th 
the first group (Samples 11, 24, 41, 54) have rough slope and As shown in Figure 3, the average, best, worst values of the site ISPRS | 
dense vegetation; the sites in the second group (Samples 12, 21, accuracy rate are 85%, 96% and 70% respectively. The standard Ne the pe 
22, 23, 31, 42) are relatively flat urban area; and the sites in the deviation of the accuracy rate and Kappa coefficient in fifteen e site a 
third group (Samples 51, 52, 53, 61, 71) contain rough terrain sites are 8% and 25%, which means the overall accuracy !5 NT Ormance 
and discontinuous (e.g., river banks and mining fields). The relatively stable while the Kappa coefficient varies depending ANT and hi 
following discussion will refer to these three groups. on the study sites. However, the parameters in the proposed iih d eee 
MTF method have to be tweaked to obtain the best results To has ; 
Figure 3 shows the quantitative assessment results of the 15 during the finite number of experiments, and the optimal result T e basi 
sites, while Table 2 lists the parameters used to generate these is not guaranteed in these experiments. In order to analysis the ype II error 
is a difficulty 
results. The overall accuracy and Kappa coefficient for one site performance of the proposed MTF method on different 
also the key t 
may be required from tests with different combination of situations, a series of comparisons are carried out as follows. 
346