Ground Segmentation Using RANSAC

This repository provides an implementation of High-Frequency Ground Segmentation using a RANSAC-based approach.
The method is designed to efficiently segment ground and obstacles from 3D point cloud data collected from LiDAR sensors.

Paper

This implementation is based on the following research:

➡ High-Frequency Ground Segmentation for Autonomous Mobile Robots: A RANSAC-Based Approach

Features

• Uses RANSAC (Random Sample Consensus) to estimate a ground plane from point cloud data.
• Applies height filtering to refine the input data before processing.
• Segments the ground and obstacles efficiently.
• Publishes the segmented point clouds to separate ROS topics for further processing.
• Designed for high-frequency real-time applications.

Installation

Prerequisites

Ensure you have the following installed:

• ROS (Robot Operating System)
• PCL (Point Cloud Library)
• C++ (with ROS CMake)

Cloning and Building the Package

cd ~/catkin_ws/src
git clone https://github.com/emirhancibir/ground_segmentation.git
cd ~/catkin_ws
catkin_make
source devel/setup.bash

Usage

Running the Node

To start the ground segmentation node:

rosrun ground_segmentation ground_segmentation_with_ransac

Or using the launch file:

roslaunch ground_segmentation ground_segmentation.launch

Parameters (can be modified in the launch file)

• distance_threshold: RANSAC inlier threshold (e.g., 0.05)
• z_threshold: Height-based filtering threshold (e.g., 0.3)
• normal_angle_threshold: Minimum Z-normal alignment for valid planes (cosine of angle)

Subscribed Topic

• /cloud_all_fields_fullframe: Input point cloud (configurable via input_topic param)

Published Topics

• /ground_points: Segmented ground points
• /nonground_points: Segmented obstacle points

Visualizing in RViz

To visualize the segmented results:

rosrun rviz rviz

Then add the following PointCloud2 display topics:

• /ground_points
• /nonground_points

Author

• Emirhan Cıbır – @emirhancibir

License

This project is licensed under the MIT License.