We dream a magic button for 3-D point cloud processing

aurivus|REVIT, 3D Studio MAX, 3D Coat, Lightwave, ZBrush, VRMesh, TopoGun, Blender, Rhino, Trimble Sketchup Etc.
Post Reply
User avatar
Joon
V.I.P Member
V.I.P Member
Posts: 329
Joined: Wed Aug 21, 2013 8:01 pm
10
Full Name: Sung Joon Ahn
Company Details: CurvSurf
Company Position Title: Founder + CEO
Country: Republic of Korea
Linkedin Profile: Yes
Location: Seongnam-si, Korea
Has thanked: 7 times
Been thanked: 32 times
Contact:

Re: We dream a magic button for 3-D point cloud processing

Post by Joon »

The secret to FindSurface's behavioral adaptability (III)

The secret is to understand what we want.

We want to extract and measure 2D/3D curves/surfaces from a 2D/3D point cloud (a set of measurement points). A curve/surface has shape, size, position, and orientation in 2D/3D space. This is the most compact and comprehensive information for describing a 2D/3D object geometry. The corresponding inlier points must be helpful for determining the bounding box (length/width/height) of the object geometry measured.

IMPORTANT: The position of an object must be defined within the body object geometry.
CurvSurf FindSurface ShapeSize6DOF.png
You do not have the required permissions to view the files attached to this post.
User avatar
Joon
V.I.P Member
V.I.P Member
Posts: 329
Joined: Wed Aug 21, 2013 8:01 pm
10
Full Name: Sung Joon Ahn
Company Details: CurvSurf
Company Position Title: Founder + CEO
Country: Republic of Korea
Linkedin Profile: Yes
Location: Seongnam-si, Korea
Has thanked: 7 times
Been thanked: 32 times
Contact:

Re: We dream a magic button for 3-D point cloud processing

Post by Joon »

The secret to FindSurface's behavioral adaptability (IV)

The secret lies in utilizing analytic geometry of curve/surface.

A curve/surface is approximately linear locally. A curve can be approximately described as line segments and a surface as triangle meshes. By analyzing local curvatures of 2D/3D point cloud data, we can assume the shapes, sizes, positions, and orientations of 2D/3D object curves/surfaces.

The assumed shapes, sizes, positions, and orientations are used as initial values for nonlinear model fitting (orthogonal distance fitting, ODF).
Why FindSurface LocalCurvatureAnalysis.png
You do not have the required permissions to view the files attached to this post.
User avatar
Joon
V.I.P Member
V.I.P Member
Posts: 329
Joined: Wed Aug 21, 2013 8:01 pm
10
Full Name: Sung Joon Ahn
Company Details: CurvSurf
Company Position Title: Founder + CEO
Country: Republic of Korea
Linkedin Profile: Yes
Location: Seongnam-si, Korea
Has thanked: 7 times
Been thanked: 32 times
Contact:

Re: We dream a magic button for 3-D point cloud processing

Post by Joon »

The secret to FindSurface's behavioral adaptability (V)

The secret is how to automate the extraction/measurement of object surfaces from point cloud data:
- Faces of box are planar.
- Cylinder is a special instance of cone with zero vertex angle.
- Sphere is a special instance of torus with zero mean radius.
This gif will help you understand what it is.
- Cylinder is a small portion of torus with infinite mean radius.

We start with finding a plane (planar), or a cone (parabolic), or a torus (elliptic/hyperbolic) according to the local surface type of point cloud data. Eventually, box/plane/sphere/cylinder/cone/torus can be automatically classified and measured.
FindSurface Automation.png
You do not have the required permissions to view the files attached to this post.
User avatar
Joon
V.I.P Member
V.I.P Member
Posts: 329
Joined: Wed Aug 21, 2013 8:01 pm
10
Full Name: Sung Joon Ahn
Company Details: CurvSurf
Company Position Title: Founder + CEO
Country: Republic of Korea
Linkedin Profile: Yes
Location: Seongnam-si, Korea
Has thanked: 7 times
Been thanked: 32 times
Contact:

Re: We dream a magic button for 3-D point cloud processing

Post by Joon »

Cylinder is a special instance of cone with zero vertex angle.

Fully automatic feature extraction from point cloud
https://youtu.be/42ZKbkjpBbY
Cylinder_Cone.gif
You do not have the required permissions to view the files attached to this post.
User avatar
Joon
V.I.P Member
V.I.P Member
Posts: 329
Joined: Wed Aug 21, 2013 8:01 pm
10
Full Name: Sung Joon Ahn
Company Details: CurvSurf
Company Position Title: Founder + CEO
Country: Republic of Korea
Linkedin Profile: Yes
Location: Seongnam-si, Korea
Has thanked: 7 times
Been thanked: 32 times
Contact:

Re: We dream a magic button for 3-D point cloud processing

Post by Joon »

Apple described its new product Vision Pro as ‘a spatial computer that blends digital content and apps into your physical space.’ We would like to look at the related functions provided by visionOS, the OS of Vision Pro, point out its shortcomings, and suggest solutions.

The framework responsible for spatial computing functions in visionOS is ARKit. Specifically, they are MeshAnchor and PlaneAnchor. MeshAnchor is mesh information about the surrounding environment obtained by processing Vision Pro's 3D measurement sensor information in real time, and PlaneAnchor is information about horizontal/vertical planes existing in the surrounding environment obtained by processing MeshAnchor.

However, it cannot be confirmed that MeshAnchor was used in the Vision Pro operation demos released so far by Apple. Apple simply displays ‘digital content and apps’ floating in the surrounding space, which is possible without using a MeshAnchor (but with using a PlaneAnchor for the office floor or table top). In this respect, Vision Pro is still a spatial monitor rather than a spatial computer (although it can represent a ball bouncing off or rolling over a MeshAnchor).

In order for Vision Pro to become a spatial computer, ‘digital content and apps’ must use the object’s geometric information represented by MeshAnchor. The essential geometric information for this is the shape, size, and 6DoF information of the object wrapped by the MeshAnchor.

CurvSurf developed the FindSurface framework to accurately extract and calculate the shape, size, and 6DoF information of objects in real time by analyzing iOS ARKit's ARMeshAnchor.
Rolling Ball - Apple iPad Pro LiDAR

youtu.be/sMRfX334blI

Types of object shapes that can be extracted and calculated include plane/sphere/cylinder/cone/ring. If their accurate size and 6DoF information can be calculated in real time, various spatial computing applications are possible.

Apple Vision Pro will be a groundbreaking spatial computer.

I hope this post is helpful, and here is an example of an application of spatial computing.
Virtual ads inside Chungmuro station Line 3 - iPhone Pro 12

youtu.be/BmKNmZCiMkw
User avatar
Joon
V.I.P Member
V.I.P Member
Posts: 329
Joined: Wed Aug 21, 2013 8:01 pm
10
Full Name: Sung Joon Ahn
Company Details: CurvSurf
Company Position Title: Founder + CEO
Country: Republic of Korea
Linkedin Profile: Yes
Location: Seongnam-si, Korea
Has thanked: 7 times
Been thanked: 32 times
Contact:

Re: We dream a magic button for 3-D point cloud processing

Post by Joon »

The secret to FindSurface's behavioral adaptability (VI)

The secret is in selecting just ONE seed point per object.

Hough transform processes every point of a scene. RANSAC selects repeatedly at least two points (2D/3D line) among the input points. Basically, they are not suitable for automatic real-time processing of large point data.

For automatic processing of point clouds in real time, the selection of only one seed point per object is mandatory. The object geometry around the seed point would then be detected and measured.
FindSurface Web Demo

The seed points could be provided from segmenting the scene image of point cloud.
User avatar
Joon
V.I.P Member
V.I.P Member
Posts: 329
Joined: Wed Aug 21, 2013 8:01 pm
10
Full Name: Sung Joon Ahn
Company Details: CurvSurf
Company Position Title: Founder + CEO
Country: Republic of Korea
Linkedin Profile: Yes
Location: Seongnam-si, Korea
Has thanked: 7 times
Been thanked: 32 times
Contact:

Re: We dream a magic button for 3-D point cloud processing

Post by Joon »

The secret to FindSurface's behavioral adaptability (VII)

The secret is to keep the area of interest minimal but sufficient.

The space is infinite, and the size of point cloud data is large.

When we human visually detect, extract, and measure a specific 3D object, we do not consider the entire environment around us.

The secret is how to define such necessary and sufficient coverage.

FindSurface does just that without using ML/DL:
FindSurface Web Demo
Post Reply

Return to “Modeling Software”