Hi Daniel!
I segmented (or split) a large point cloud (100 million points) into 10 segments (8 to 12 millions each) to avoid crashes during CANUPO and M3C2 computations, and then merged them back together. I'm not sure if cloudcompare employs a buffer during segmentation or splitting, so I was wondering how accurate was the merge process? like did segmentation and merge processes altered the geometry of the point cloud or did I lose any points at the boundaries?
Regards!
Precision of Segment and Merge Operations
Re: Precision of Segment and Merge Operations
The merge process is a very simple concatenation of point clouds. There's no 'spatial' merging. At least it's lossless.
Daniel, CloudCompare admin
Re: Precision of Segment and Merge Operations
Hi Daniel!
1) Please look at the first Figure below and notice the three horizontal lines. That's where I segmented my point cloud (into 3 parts) for classification. After classification when I merged the three parts back together I got these lines representing vegetation. Is there a way to avoid these lines?
2) Also,when I split the classified point cloud into ground points and vegetation using scalar field, I see that some of the points especially at the lower right corner were included in both ground points as well as vegetation (when displayed in RGB). (Figure 2 below).
3) How about the grey points (that were not classified as ground or vegetation)? Are they included in both ground points and vegetation clouds, when original point cloud is split by specifying the corresponding scalar fields?
1) Please look at the first Figure below and notice the three horizontal lines. That's where I segmented my point cloud (into 3 parts) for classification. After classification when I merged the three parts back together I got these lines representing vegetation. Is there a way to avoid these lines?
- point cloud classified. .jpeg (208.1 KiB) Viewed 3637 times
- grey points.jpg (115.11 KiB) Viewed 3637 times
Re: Precision of Segment and Merge Operations
1) These lines may be due to border effects of the classification process I guess? If yes I don't know how to avoid this.
2) Are they the exact same points? Or points very close to each other?
3) Grey points are probably points that could not be classified in any class? How have you generated these colored clouds? I'm not sure to understand what I'm looking at :D
2) Are they the exact same points? Or points very close to each other?
3) Grey points are probably points that could not be classified in any class? How have you generated these colored clouds? I'm not sure to understand what I'm looking at :D
Daniel, CloudCompare admin
Re: Precision of Segment and Merge Operations
1) Yea, these are certainly the borderline effects. I'm wondering if these are due to edge artifacts induced by segmentation and merging without employing a suitable buffer? Is there an option in cloud compare to segment point clouds or surfaces with buffer?
2) They are the exact same points. In second figure, the right most is the classified point cloud (both ground and veg) displayed as scalar fields, middle one is the vegetation exported from the right most figure and displayed as RGB, and the left one is the ground points (bare-earth) displayed as RGB. You might get confused because there're green colored points in the bare-earth model (left part of figure 2), but that is actually very fine grass/moss on rock surface. I'm okay with that classified as ground. In fact I actually wanted that.
3) it's a photogrammetric point cloud.
2) They are the exact same points. In second figure, the right most is the classified point cloud (both ground and veg) displayed as scalar fields, middle one is the vegetation exported from the right most figure and displayed as RGB, and the left one is the ground points (bare-earth) displayed as RGB. You might get confused because there're green colored points in the bare-earth model (left part of figure 2), but that is actually very fine grass/moss on rock surface. I'm okay with that classified as ground. In fact I actually wanted that.
3) it's a photogrammetric point cloud.