Page 15 - 中国仿真学会通讯2020第1期
P. 15
are used together to ensure the integrity of the Algorithm 1:The Algorithm to Layer the Facade Point
building model. In order to solve the problem of Cloud
complexity and fidelity, we propose an optimized
modeling method based on hierarchical Input: The facade point cloud of the building
reconstruction. This method uses plane as the
basic unit. On the one hand, it can make up for 1. Using the P⁃Linkage algorithm to extract and segment
the modeling missing problems and flatten the the facade point cloud, which results in the point
surface of the building and improve the fidelity of cloud being divided into planar patches
the building model. On the other hand, using the
plane as the basic unit can greatly simplify the 2. Extracting the main direction of the building
model and reduce the complexity of the model. for i = 1 to m ( m represents the number of planar
Therefore, how to hierarchically segment and patches)
model the building point cloud are the core issue calculating the normal vector n of each planar
of this paper. patch PPi Clustering planar patches according to the
normal vector nCounting the number of blocks of
Algorithm 1 presents the method to layer the each cluster
facade point cloud. The algorithm extracts and Extracting the three normal vector with the largest
clusters the planar patches by P⁃Linkage [ 16 ] number of planar patches as the main direction( the
algorithm to get the three main directions of the
building. Based on the normal vector and the front and back direction →nfb , the left and right
spatial positional relationship of the planar patches direction →nlr , the upper and lower direction n→ud ) ,
and layered the point cloud of the building facade.
The algorithm can automatically extract three main which are shown in Figure 3
directions and hierarchically extract the point 3. for i = 1 to m
cloud of the building facade building that satisfies
the Manhattan hypothesis. Algorithm 2 c(i) = index ( from 1 to 3 ) of cluster whose
reconstructs the building facade based on the difference from the three main direction satisfies
layered point cloud obtained by Algorithm 1. By the threshold
converting the layered point cloud into a black
and white picture and extracting the wire frame, for i = 1 to m
the layered point cloud can be adaptively Clustering planar patches into the front and back
converted into a wire frame. The method is simple
and easy to understand and the model built by the layered point cloud Pfb ( c(i) = 1 ) , the left and
method is extremely low complexity. Algorithm 1, right layered point cloud Plr ( c(i) = 2 ) and the
Algorithm 2 and Algorithm 3 will be introduced in upper and lower layered point cloud Pud( c(i) ) =
3.1, 3.2 and 3.3 respectively. 3) , as shown in Figure 4( b)
3.1 The Layered Algorithm 4. Rotating the main direction of the front and back
For the facade point cloud, we use the layers →nfb to the X⁃axis ( 1,0,0) direction based on
Algorithm 1 to segment the facade point cloud into
planar patches, and cluster the planar patches to equation ( 1 ) , and getting the three new main
layer the facade point cloud.
directions n→FB , n→LR , and n→UD , which are shown in
12
Figure 5.
5. Extracting the front and back layered point cloud Pfb
( Figure 4( c) ) and projecting it onto the XOZ plane
to get the point cloud collection P (fb⁃lr Figure 4( d) )
Figure 3: The Three Main Directions of the
Facade Point Cloud.