Ray Casting#

The RaycastingScene class in Open3D provides basic ray casting functionality. In this tutorial we show how to create a scene and do ray intersection tests. You can also use RaycastingScene to create a virtual point cloud from a mesh, such as from a CAD model.


As the first step we initialize a RaycastingScene with one or more triangle meshes.

# Load mesh and convert to open3d.t.geometry.TriangleMesh
cube = o3d.geometry.TriangleMesh.create_box().translate([0, 0, 0])
cube = o3d.t.geometry.TriangleMesh.from_legacy(cube)
# Create a scene and add the triangle mesh
scene = o3d.t.geometry.RaycastingScene()
cube_id = scene.add_triangles(cube)

add_triangles() returns the ID for the added geometry. This ID can be used to identify which mesh is hit by a ray.


Casting rays

We can now generate rays which are 6D vectors with origin and direction.

# We create two rays:
# The first ray starts at (0.5,0.5,10) and has direction (0,0,-1).
# The second ray start at (-1,-1,-1) and has direction (0,0,-1).
rays = o3d.core.Tensor([[0.5, 0.5, 10, 0, 0, -1], [-1, -1, -1, 0, 0, -1]],

ans = scene.cast_rays(rays)

The result contains information about a possible intersection with the geometry in the scene.

dict_keys(['primitive_uvs', 'primitive_ids', 'geometry_ids', 'primitive_normals', 't_hit'])
  • t_hit is the distance to the intersection. The unit is defined by the length of the ray direction. If there is no intersection this is inf

  • geometry_ids gives the id of the geometry hit by the ray. If no geometry was hit this is RaycastingScene.INVALID_ID

  • primitive_ids is the triangle index of the triangle that was hit or RaycastingScene.INVALID_ID

  • primitive_uvs is the barycentric coordinates of the intersection point within the triangle.

  • primitive_normals is the normal of the hit triangle.

We can see from t_hit and geometry_ids that the first ray did hit the mesh but the second ray missed.

print(ans['t_hit'].numpy(), ans['geometry_ids'].numpy())
[8.999999      inf] [         0 4294967295]

Creating images

We now create a scene with multiple objects

# Create meshes and convert to open3d.t.geometry.TriangleMesh
cube = o3d.geometry.TriangleMesh.create_box().translate([0, 0, 0])
cube = o3d.t.geometry.TriangleMesh.from_legacy(cube)
torus = o3d.geometry.TriangleMesh.create_torus().translate([0, 0, 2])
torus = o3d.t.geometry.TriangleMesh.from_legacy(torus)
sphere = o3d.geometry.TriangleMesh.create_sphere(radius=0.5).translate(
    [1, 2, 3])
sphere = o3d.t.geometry.TriangleMesh.from_legacy(sphere)

scene = o3d.t.geometry.RaycastingScene()
_ = scene.add_triangles(sphere)

RaycastingScene allows to organize rays with an arbitrary number of leading dimensions. For instance we can generate an array with shape [h,w,6] to organize rays for creating an image. The class also provides helper functions for creating rays for a pinhole camera. The following creates rays Tensor with shape [480,640,6].

rays = o3d.t.geometry.RaycastingScene.create_rays_pinhole(
    center=[0, 0, 2],
    eye=[2, 3, 0],
    up=[0, 1, 0],
# We can directly pass the rays tensor to the cast_rays function.
ans = scene.cast_rays(rays)

The output tensors preserve the shape of the rays and we can directly visualize the hit distance with matplotlib to get a depth map.

import matplotlib.pyplot as plt
<matplotlib.image.AxesImage at 0x7fd6bc81bd00>

Further we can plot the other results to visualize the primitive normals, ..

# use abs to avoid negative values
<matplotlib.image.AxesImage at 0x7fd6b8bf68b0>
plt.imshow(ans['geometry_ids'].numpy(), vmax=3)
<matplotlib.image.AxesImage at 0x7fd6b8ba7280>

Creating a virtual point cloud#

We can also use the hit distance to calculate the XYZ coordinates of the intersection points. These are the points that you would get by placing a virtual 3D sensor at the point of origin of the rays.

hit = ans['t_hit'].isfinite()
points = rays[hit][:,:3] + rays[hit][:,3:]*ans['t_hit'][hit].reshape((-1,1))
pcd = o3d.t.geometry.PointCloud(points)
# Press Ctrl/Cmd-C in the visualization window to copy the current viewpoint
                                  front=[0.5, 0.86, 0.125],
                                  lookat=[0.23, 0.5, 2],
                                  up=[-0.63, 0.45, -0.63],
# o3d.visualization.draw([pcd]) # new API