koide3/gtsam_points

A collection of GTSAM factors and optimizers for point cloud SLAM

What it solves

gtsam_points provides a specialized set of tools for range-based SLAM (Simultaneous Localization and Mapping), specifically focusing on the registration and optimization of 3D point clouds. It extends the GTSAM library to enable more accurate and efficient point cloud matching, trajectory estimation, and mapping.

How it works

The project implements several key components for point cloud processing:

  • Scan Matching Factors: Implements various Iterative Closest Point (ICP) variants, including point-to-point, point-to-plane, Generalized ICP (GICP), and Voxelized GICP (VGICP) with GPU acceleration.
  • Colored Scan Matching: Integrates photometric error into the matching process to improve registration using color information.
  • Continuous-time ICP: Provides factors for continuous-time trajectory estimation, reducing the motion blur effect in LiDAR scans.
  • Bundle Adjustment: Implements factors based on eigenvalue minimization (EVM) and least-squares for multiview point cloud registration.
  • Nearest Neighbor Search: Includes high-performance search structures like KdTree and Incremental Voxel Maps (iVox) for fast data association.
  • Global Registration: Uses RANSAC and Graduated Non-Convexity (GNC) to align point clouds without a good initial guess.
  • Segmentation: Offers region growing and min-cut algorithms to partition point clouds into objects.

Who it’s for

Robotics engineers and researchers working on LiDAR-based SLAM, 3D reconstruction, and autonomous navigation systems that rely on the GTSAM framework.

Highlights

  • GPU Acceleration: Includes a GPU implementation of VGICP for faster processing.
  • Cubic B-Splines: Supports continuous-time trajectory representation for smoother motion estimation.
  • Extensive Factor Library: A wide range of matching costs (LOAM, GICP, Colored ICP) integrated directly into a factor graph optimization framework.
  • Efficient Data Structures: Features incremental voxel mapping and fast occupancy grids for efficient point cloud overlap estimation.

Related

  • Project
    koide3/glim

    A versatile 3D mapping framework that uses factor graphs and GPU acceleration to provide accurate, sensor-agnostic range-based mapping and localization.

  • Project
    borglab/gtsam

    A C++ library for smoothing and mapping in robotics and vision that uses factor graphs and Bayes networks for efficient state estimation and optimization.

  • Project
    koide3/small_gicp

    A highly optimized, parallelized C++ library for fine point cloud registration (ICP, GICP, VGICP) designed for high-speed 3D alignment.

  • Project
    SJTU-ViSYS/Ground-Fusion

    A low-cost sensor-fusion SLAM system for ground vehicles that integrates RGB-D, IMU, wheel odometry, and GNSS to maintain robust localization during sensor failures.

  • Project
    norlab-ulaval/libpointmatcher

    A modular C++ library with Python bindings that implements the Iterative Closest Point (ICP) algorithm for aligning 3D point clouds in robotics and computer vision.