CANUPO (Multiscale Dimensional Classification) is one of the binary point cloud classification methods added in CloudCompare.

It is a technique used across point cloud data to reduce the dimensionality of the data while preserving the original features.

The idea of this classification method is to extract features at multiple scales, which can capture both fine and coarse details of the entire point cloud. This method particularly useful in complex environments where objects and structures are in different scale, shape and dimension.

Steps to the CANUPO classification

1.Segment the target features manually

2.Use the CANUPO training window and input the parameters and segmented target features for the classification process

3.View the training result, load the statistics and output the .prm file

4.Use the CANUPO classification window and input the generated .prm file

5.Wait until the classification point cloud process is complete

Introducing some simple tree visualization methods for point cloud data, including shading, stereogram and voxelizaiton. 

The shading can give a brief way to visulised the point cloud by sunlight simulation and shadow analysis, the canopy layer, tree trunks and leafs can be simulated with the sun angle. 

Stereogram can effective shows the leaf and tree truck inclination and tilting angle by analysing its scanned structure. Structural defects can be detected with stereogram and the phototropism can be also eveulated by this graph. 

Nevertheless, we do the voxelization with Cloudcompare, this process can makes the unstructured point cloud into a structural voxel (3d pixel).

Meshing and modelling with different point cloud data

Airborne LiDAR roof-based meshing and upscaling point cloud

•Manually clipping and trimming of the target building 

•Triangulation with the original point cloud

•Meshed result to point cloud

Close-ranging trunk-based skeletonization

•Extract and convert the tree point cloud to array

•Use the DBSCAN clustering for the node extraction

•Set the node into different labels

•Build a Kd tree for the nearest neighbor searching

•Skeletonization from the Kd tree result

•Convert the trunk skeleton into a new point cloud

•Visualized the point cloud

Basic point cloud processing technqiues with Cloudcompare, including: 

1. CSF simulations (filter the ground and non-ground points) 

2. Noise filtering

3. Segementation by color 

4. Point normal computation 

5. Voxelization and re-interpolate with colors 

6. Airborne LiDAR pre-clasified vegetation segmentation 

7. Leaf area index by voxelization 

There introduced techniques can be applied arocess both Airborne and close-ranging point cloud datasets.