neka-nat/probreg

What it solves

It provides a more robust alternative to the Iterative Closest Point (ICP) algorithm for point cloud registration. By using probabilistic models, it can handle noise and complex transformations more effectively than traditional methods.

How it works

The library implements several stochastic registration algorithms that align two point sets by treating them as probabilistic distributions. It supports both rigid transformations (like 6D pose and scale) and non-rigid transformations (such as Affine, MCT, and TPS).

Who it’s for

Researchers and developers working with 3D point clouds, computer vision, and robotics who need reliable alignment tools that integrate with Open3D.

Highlights

• Diverse Algorithm Suite: Includes Coherent Point Drift (CPD), FilterReg, GMMReg, and Support Vector Registration (SVR).
• Open3D Integration: Provides a simple interface for working with Open3D point cloud objects.
• Flexible Transformations: Supports rigid, affine, and non-rigid (deformable) registration.
• CUDA Support: Accelerates Rigid and Affine CPD using CuPy.