Hello,
I’m working on my master thesis and am currently searching for efficient ways to mesh a very large pointcloud (Lidar: 2x10^9 points), so that I can diplay the geological site as a walkable landscape in Unity.
RealityCapture does a very neat job of creating goodlooking meshes and the simplify tool streamlines the editing process quite a bit.
My plan is to rasterize the landscape in mesh-cubes of about 15x15x15 m. Each mesh will have several levels of detail to make the visualization possible on weaker PCs.
Now I’ve encountered two problems with this in RealityCapture and was hoping you might have a hint on how to get around them.
-
When creating two meshes from adjascent reconstruction regions I always get a small seam at the edges (see first picture).
I can get rid of the seam by
a) creating a larger reconstruction area than necessary
b) creating a mesh from this
c) making the reconstruction region smaller until it fits
d) filtering the unnecessary vertices
c) texturizing and exporting the mesh
By exporting, editing and importing reconstruction regions I get very precise results but it takes a lot of time. So, is there a way to avoid the seam by making changes in the settings?
-
RealityCapture in the Steam version doesn’t seem to allow for an automated tiling of the point cloud and even the CLI version doesn’t seem to include commands that would make it possible to create a batch process that creates mesh-tiles in a specified grid.
One way around this would be to create a very large scale mesh and save it by parts. This is actually a pretty comfortable way of getting it done - BUT I don’t have a lot of control over the individual parts (texture creation, level of detail, exact coordinates for recalculation).
The only way around this I could think of is to create a large number of files (around 2000) with reconstruction regions forming the grid and import them one after the other by hand.
So, did I maybe miss an easy way to get this done or will there be a feature like this in the very near future? Ideally including a solution for the seam in 1?
Hi Linus
Regarding the seam:
I can see misalignments in the image ( RED-circled ) so it is highly probable the crack is visible there because of this.
Terrain_crack.jpg
Normally you will not see any cracks as model parts are perfectly reconstructed per triangle side. You can check it in other parts of the model, where parts are connected together perfectly…
Regarding cutting into smaller pieces, for now it is not possible to cut into custom parts ( could be alternatively possible with the reconstruction region and FILTERING tool and creating a bounding box of a predefined size and then filter parts )
But as RC creates parts based on estimated vertices-triangle counts, then it could be even better to use parts as they come directly from RC and then calculate normals, displacement and etc maps in some external apps…
Thanks for the reply Wishgranter
Wishgranter wrote:
Hi Linus
Regarding the seam:
I can see misalignments in the image ( RED-circled ) so it is highly probable the crack is visible there because of this.
Normally you will not see any cracks as model parts are perfectly reconstructed per triangle side. You can check it in other parts of the model, where parts are connected together perfectly…
Yes, the “mesh-parts” created by RC fit together without a seam, but the borders of the whole mesh (where the reconstruction area cuts it off) always look a bit round and often show some artifacts. So this is only a problem I encounter when I manually divide the mesh with reconstruction regions.
Wishgranter wrote:
But as RC creates parts based on estimated vertices-triangle counts, then it could be even better to use parts as they come directly from RC and then calculate normals, displacement and etc maps in some external apps…
I am looking into this. Creating mesh-tiles from reconstruction regions has the advantage that I can comfortably use your very potent “simplify tool” to adjust the indiviual parts according to their individual level of detail. Additionally I get a very organised filestructure with evenly sized cubes that allows for easier editing of specific areas in my landscape. I could even go back into RC and reexport a single cube as long a I have the coordinates and size of the reconstruction region.
Would it be possible with the CLI licence to import a reconstruction region, do-some-meshing-and-editing and then load the next reconstruction region to do the same?
Because then I could create the necessary textfiles in batch and let RC work its way through them.
For example in the attached simple grid. Could I load the reconstruction region X.rcbox
and then let RC go through a simple loop setting X=1 and adding 1 until it gets to X=9?
snip_20161220161428.png