Point cloud generation and preprocessing result

After searching and implementing various methode to modify the mesh before applying KSR, we tested it on the meshes presented in the Software and Data section. We can compare the result with the result we got few months ago during the project on the same subject.

We can take a look at what the preprocessing turned out, below are the point clouds generated from the workflow applied to the Three Zone meshes:

Table 1. Point cloud generated from Three_zones

Polygon Mesh from the STL files

Point Cloud generated

Normal associated

3zones stl

3zones ply

normal 3zones

From the Stl files we successfully create a dense point cloud with normal associated outwardly oriented.Now we can try to use it in KSR. Based on the same Mesh, few months ago we tried to generated the point cloud generate with a software cloodcompare,we can compare both outcome and see the improvements:

Table 2. comparaison of Kinetic outcome

Old KSR outcome

New KSR outcome

3zones v2 1

3zones final

The old result had degenerated faces, Although it gave an idea of the shape and what he tries to generate but it was not functional.But in comparaison, now we got a good representation of the mesh with the exception of windows who are replaced by empty space.

To generate precise structure, we had to use low value as parameters for the minimum region size, generating more detailes face. For instance for the outcome from before we use the following parameters : - min.region.size=150 - maximum.distance=0.06

Different parameters lead to different result :

Table 3. comparaison of different parameters execution

Parameters used :

min.region.size = 500 ; maximum.distance 0.06

min.region.size = 45 ;maximum.distance =0.06

KSR output

3zone 500 0.06

3zones 45 0.06

Mesh

3zones 500 0.06mesh

3zones 45 0.06 mesh

Increasing the min.region.size parameters result in less details, but a simpler mesh, in this case we got a parallelepiped with two wall inside it, and if we decrease even more we start to getting the shape of the windows or door but it does not succeed in filling completly the hole.For simulation the parallelepiped might be better to ensure the watertightness of the model.

Here’s an exemple with another mesh :

Table 4. ACJasmin example

ACjasmin mesh STL

ACjasmin point cloud

ACjasmin KSR outcome

ACjasmin KSR mesh

jasmin stl

jasmin ply

Image 3

Image 4

We end up loosing some details like windows, but the mesh is supposely closed, watertight. The distinction of the different room without doors is still preserved.

We can also tried with more difficult mesh like this dragon found on footnote:https://free3d.com/fr/3d-model/black-dragon-rigged-and-game-ready-92023.html[free 3d],

Table 5. Example with more complex mesh

Dragon stl

Dragon from KSR

superposition of both

dragon

dragon KSR

both dragon

The general shape of the Dragon is kept and understandable but the head is reduce to a cube without horns,eyes or all small details. With both of the last examples, we can see that our workflow works well on low-detail meshes but has difficulties handling highly detailed meshes with a lot of curvature or very small objects like scales, door handles, or window corners.