CurvSurf, Inc.

24/12/2025

An experimental sample project for Galaxy XR has been developed, which records and filters point clouds reconstructed from ARCore depth maps (Android XR). You can find the project on GitHub: https://github.com/CurvSurf/PointCloudRecorder-GalaxyXR .

Currently, Android XR exposes depth data suitable for visualization; however, it lacks sufficient synchronized and calibrated information for reliable metric point cloud reconstruction.

As a result, we have decided to abandon further attempts at integrating FindSurface on this project and instead publish our findings here.

An experimental sample project for Galaxy XR that records and filters point clouds reconstructed from ARCore depth maps (Android XR). - CurvSurf/PointCloudRecorder-GalaxyXR

04/11/2025

Within the next two months, CurvSurf will release the source code for its Galaxy XR app, similar to the Vision Pro app:
https://youtu.be/2aSMBrPTEtg

Follow the project:
https://github.com/CurvSurf/FindSurface

FindSurface library's main hub for various platforms/languages. - CurvSurf/FindSurface

28/09/2025

CurvSurf FindSurface leverages Apple's ARPointCloud (rawFeaturePoints) to extract and measure object geometries. The ARPointCloud serves as a debug visualization for the tracking algorithm, yet CurvSurf effectively utilizes the data it provides for its operations. Watch more about this integration here: https://youtu.be/SIdQRiLj2jY

CurvSurf FindSurface is extracting and measuring object geometries from Apple ARPointCloud (rawFeaturePoints)https://developer.apple.com/documentation/arkit/...

03/09/2025

Combining the two approaches FindSurface and ML/DL

The following questions and answers using Gemini or Copilot demonstrate a significant advance in the processing and analysis of 3D point cloud data.

Q1: What are the difficulties in obtaining geometric object information by analyzing and processing a 3D point cloud with ML/DL?
A1: ...

Q2: Compare the methods for obtaining geometric object information by analyzing and processing a 3D point cloud using ML/DL and CurvSurf's FindSurface.
A2: ...

Q3: What if both methods were combined, e.g., applying FindSurface to segmented points through ML/DL? Again, ML/DL leverages the object information obtained from FindSurface.
A3: Combining ML/DL and geometry-based methods like CurvSurf's FindSurface is not just a theoretical possibility; it's an active and promising area of research and development in 3D point cloud analysis. This hybrid approach seeks to leverage the strengths of each method while mitigating their weaknesses, leading to a more robust, accurate, and versatile pipeline.

Here's how these two approaches can be combined and the benefits and challenges of doing so:

Scenario 1: ML/DL Guides FindSurface (The Common Hybrid Pipeline)
This is the most direct and widely explored combination. An ML/DL model acts as a "pre-processor" for the geometry-based algorithm.

Scenario 2: FindSurface Feeds ML/DL (The Feature Engineering Approach)
In this less common but equally powerful scenario, the geometric information from FindSurface is used to enrich the input features for an ML/DL model.

Scenario 3: Feedback loop between 1 & 2.

Send a message to learn more

22/08/2025

The FindSurface runtime library (SDK) takes a unique approach by avoiding traditional algorithms like k-d trees (octrees), RANSAC, and the Hough transform due to their limitations in real-time 3D data processing.

- **k-d trees (octrees):** While k-d trees offer efficient point finding at O(log n) for large point clouds, the time-intensive process lies in structuring the tree at O(n log n) from the incoming cloud.
- **RANSAC (Hough transform):** These methods excel in fitting lines/planes with over 50% inliers but face exponential complexity growth with additional parameters or input points.

FindSurface stands out by locally searching for object geometry within the point cloud, eliminating the need for preprocessing such as meshing, smoothing, or creating k-d tree (octree) structures for the entire dataset.

Explore more about FindSurface's innovative approach

FindSurface library's main hub for various platforms/languages. - CurvSurf/FindSurface

29/07/2025

The CurvSurf's FindSurface algorithm features "Auto Detection," enabling it to identify the most suitable surface type from five options (plane, sphere, cylinder, cone, and torus) in the vicinity of the seed area without manual designation. The algorithm automatically determines the resulting surface type as the surface model converges to a specific type.

The algorithm leverages the concept that locally non-singular surfaces exhibit distinct local curvature types:
- Planar: Attempt to fit a plane.
- Parabolic: Attempt to fit a cone.
- Elliptic: Attempt to fit a torus.
- Hyperbolic: Attempt to fit a torus.

Additionally, the algorithm performs conversions as follows:
- Transforms a cone (conical frustum) with identical radii at the top and bottom into a cylinder.
- Converts a degenerate torus (a double-covered sphere) with a mean radius of zero into a sphere.
- Adjusts a torus with an infinite mean radius into a cylinder.
- Notes that a box comprises at least two orthogonal planes.

For visual reference, you can view Figure 1 at:

FindSurface library's main hub for various platforms/languages. - CurvSurf/FindSurface

16/07/2025

Let's imagine how the mean radius, tube radius, position, and orientation of an industrial pipe elbow can be approximately determined by analyzing the measurement points of a small area (seed point and seed radius, i.e. patch). This is done using local curvature analysis (differential geometry, differential calculus, statistics, and probability theory).
After the approximate estimation of the initial parameter values, then region growing and orthogonal distance fitting take over iteratively. Finally, the geometric parameter values are refined and the inlier/outlier points are separated.
Ultimately, the exact results are displayed and applied. On Apple Vision Pro 2024, 500 Hz for planes, minimum 100 Hz for spheres, cylinders, cones, and tori.
It's a magic!

Send a message to learn more

11/07/2025

CurvSurf is excited to announce the seamless integration of FindSurface into your applications.

Discover the remarkable features of FindSurface:
- Effortless one-click operation with a seed point
- Automatic identification of planes, spheres, cylinders, cones, and tori
- Real-time detection achieving rapid geometrical analysis (2–10 ms per geometry)
- Exceptional precision meeting ISO 10360-6 standards
- Compatibility with noisy, incomplete point clouds
- Compact size with an incredibly small footprint (

Send a message to learn more

06/07/2025

There are distinct disparities between seeing, understanding, and measuring.
When applied practically, these distinctions become more evident and noteworthy.
Observation is where we begin.
Comprehending what we observe is a separate skill.
Moreover, achieving precise measurements surpasses our abilities to sense and comprehend.
The task of measurement is entrusted to mathematics and specialized tools.

Real Time 3-D Object Recognition

- OS: Windows 10- CPU: Intel Core i7-6500U @ 2.5GHz- Depth Camera: Intel RealSense SR300- Middleware: CurvSurf FindSurface.FindSurface is working in single-t...

03/05/2025

3D Box Measurement
The size and orientation of 3D box are determined in real time.
DepthMap (60 Hz) data from the Apple LiDAR 3D camera is analyzed.
A further development involves image processing of the video.

3D Box MeasurementThe size and orientation of 3D box are determined in real time.DepthMap (60 Hz) data from the Apple LiDAR 3D camera is analyzed.A further d...

08/03/2025

Virtual displays on real surfaces (Apple Vision Pro).

18/10/2024

🔍 Exciting news for developers! 📱 The FindSurface real-time App for Apple Vision Pro's source code is now available on GitHub. Dive into the details and explore the functionalities of this cutting-edge app designed for Apple Vision Pro devices running on visionOS 2.0+.

⚙️ This innovative app, requiring user permissions for hand tracking and environment scanning, offers unparalleled performance:
- Plane fitting of 600 Hz
- Sphere/cylinder fitting of 300 Hz
- Cone/ring fitting of 100 Hz

Check out the GitHub link below to access the source code and start exploring the possibilities of this powerful real-time app! 👇

[GitHub Link]( https://github.com/CurvSurf/FindSurface-RealityKit-visionOS-Real-Time )

A sample project for visionOS showcasing FindSurface's real-time searching capabilities. - CurvSurf/FindSurface-RealityKit-visionOS-Real-Time