Open3d learning note 4 [surface reconstruction]

Alpha shapes

Reconstruction with convex hull .

import open3d as o3d

mesh_ply = o3d.io.read_triangle_mesh("mode/bunny.ply")
#  sampling 
pcd = mesh_ply.sample_points_poisson_disk(750)
#  Take a look at the convex hull 
hull, _ = pcd.compute_convex_hull()
hull_ls = o3d.geometry.LineSet.create_from_triangle_mesh(hull)
hull_ls.paint_uniform_color((1, 0, 0))
o3d.visualization.draw_geometries([pcd, hull_ls], width=1280, height=720)
# Start surface reconstruction ,alpha Adjustable value 
alpha = 0.1
print(f"alpha={alpha:.3f}")
mesh = o3d.geometry.TriangleMesh.create_from_point_cloud_alpha_shape(pcd, alpha)
mesh.compute_vertex_normals()
print(mesh)
o3d.visualization.draw_geometries([mesh], mesh_show_back_face=True, width=1280, height=720)

result ：

Convex hull condition

Reconstruction effect ,Alpha=0.1
continue ,Alpha = 0.015

Rolling ball method

import open3d as o3d

mesh = o3d.io.read_triangle_mesh("mode/bunny.ply")
mesh.compute_vertex_normals()
# Old rules , Sample first 
pcd = mesh.sample_points_poisson_disk(3000)
# Here you can see the normal 
o3d.visualization.draw_geometries([pcd], point_show_normal=True, width=1280, height=720)
# Start rebuilding 
radii = [0.005, 0.01, 0.02, 0.04]
rec_mesh = o3d.geometry.TriangleMesh.create_from_point_cloud_ball_pivoting(pcd, o3d.utility.DoubleVector(radii))
o3d.visualization.draw_geometries([pcd, rec_mesh], width=1280, height=720)

effect ：

Normal annotation

Reconstruction effect

Poisson surface reconstruction

The two above , The rolling ball method looks good , But in fact, both are chicken ribs , Is not smooth . Poisson reconstruction will be better in this regard .
There is only one adjustable parameter for Poisson surface reconstruction , That's it depth, This defines the depth of the octree , If the value is high, there will be more details .

import open3d as o3d

mesh = o3d.io.read_triangle_mesh("mode/Fantasy Dragon.ply")
mesh.compute_vertex_normals()
pcd = mesh.sample_points_poisson_disk(3000)

with o3d.utility.VerbosityContextManager(o3d.utility.VerbosityLevel.Debug) as cm:
    mesh, densities = o3d.geometry.TriangleMesh.create_from_point_cloud_poisson(pcd, depth=9)
# Paint him a color 
mesh.paint_uniform_color([1, 0, 0])
o3d.visualization.draw_geometries([mesh], zoom=0.664,
                                  front=[-0.4761, -0.4698, -0.7434],
                                  lookat=[1.8900, 3.2596, 0.9284],
                                  up=[0.2304, -0.8825, 0.4101],
                                  width=1280,
                                  height=720)

Next, let's go through the process , Look at the density of points .

print('visualize densities')
densities = np.asarray(densities)
density_colors = plt.get_cmap('plasma')(
    (densities - densities.min()) / (densities.max() - densities.min()))
density_colors = density_colors[:, :3]
density_mesh = o3d.geometry.TriangleMesh()
density_mesh.vertices = mesh.vertices
density_mesh.triangles = mesh.triangles
density_mesh.triangle_normals = mesh.triangle_normals
density_mesh.vertex_colors = o3d.utility.Vector3dVector(density_colors)
o3d.visualization.draw_geometries([density_mesh], zoom=0.664,
                                  front=[-0.4761, -0.4698, -0.7434],
                                  lookat=[1.8900, 3.2596, 0.9284],
                                  up=[0.2304, -0.8825, 0.4101],
                                  width=1280,
                                  height=720)

Purple is low density , Yellow is high density , You can screen out some with low density . give the result as follows ：

There seems to be no change . Sampling point 3000 In the case of reconstruction , There are still some scattered noises . To remove noise , There are two ways ： The first one is ： Continue to reduce sampling points ; The second kind , Increase sampling point . Reducing sampling points will lose more details , So it's not recommended .
Let's see the effect of increasing sampling points .

Sampling point =10000

Almost restored the original appearance , But what if the low-density area is removed ？

Become incomplete . So use the delete function depending on the situation .

Let's take a look at reducing sampling points .

Sampling point =1000

As expected , Although the original scattered points are gone , But at the same time, many details have been lost , Do more harm than good .
Next, change the model to see the reconstruction effect .

Sampling point 15000
Close to the limit , Even the sampling points basically cover the whole model , Some details will still be lost .