The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bib. Beijing 2008 
580 
Legend 
0 - .07 
.07-.13 
.13-.20 
.20 - .26 
.26 - .33 
.33 - .39 
.39 - .46 
.46 - .52 
.52 - .59 
Figure 6. Kernel density of FCD 
From figure 5, it is obvious that at each road intersection FCD 
are not distributed evenly. In the centre of road intersections, 
there are less FCD but more on the road near the centre. This 
does make sense because in the really traffic situation, due to 
the effect of traffic lights, vehicles must wait till they are 
notified to go which will cause the effect that more vehicles are 
on the road near the centre. When vehicles are allowed to go, 
they must go through road intersections without stop, in which 
situation there should be less vehicles in the centres of road 
intersections. Based on the interest area in figure 5, a kernel 
density analysis with a search radius 15M and output cell size 
4.5M is carried out and the result is shown in figure 6. 
The unit of the classification in figure 6 is number of points per 
square meters. From the analysis of the kernel density of the 
whole study area, the weight matrix is defined in equation (3), 
f- 0.4 dist(i, j) < 10m 
W u = < (3) 
[1 10 < dist(i,j) < 30m 
Where, dist(i,j) is the distance from point/ to point j . To 
be more intuitive, it is easy to explain equation (3) in the real 
traffic situation. In Shenzhen city, centred at centres of each 
road intersection, the average radius of all circles which can 
cover each intersection is 10 meter. Therefore, according to the 
analysis of kernel density shown in figure 6, to detect road 
intersections, a negative weight is preferred to give to points 
within 10 meter distance around each interest point. Along 
roads near to centres of each road intersection, the average 
waiting line of vehicle is 20 meter. Thus, according to the 
kernel density analysis, a positive weight needs to be assigned 
to points within 30 meter distance but farther than 10 meter 
distance around each interest point. Points farther than 30 meter 
distance around each interest point have little influence on 
detecting road intersections and they are not taken into account. 
5.3 Monte Carlo simulation process 
Since rules to build the weight matrix is set, the Monte Carlo 
simulation process can be carried out. There are three steps for 
each time of the simulation process. They are, 
a) 21935 points are randomly distributed in the study 
area. 
b) weight matrix is build with equation (3) 
c) the local statistic L f is calculated with equation 
(2) 
Repeat these three steps 1000 times and a series of L t can be 
found. In figure 7, results are arranged in a histogram.