Real-Time Wall Detection from 3D LiDAR Data in ROS

This repository is for real-time wall detection using 3D LiDAR data on autonomous mobile robots. It subscribes to a Velodyne VLP-16 LiDAR point cloud (/velodyne_points), extracts vertical planar surfaces (walls) via PCL filters and RANSAC segmentation, and publishes wall point clouds for navigation. Planar fitting is a common technique to detect surfaces like walls and floors in point clouds. In practice the Velodyne VLP-16 driver publishes a sensor_msgs/PointCloud2 on /velodyne_points, which wall_detect_node.cpp processes to find walls. The complementary wall_compress_node.cpp then projects these walls onto the ground plane and downsamples them for compact mapping.

• Raw pointcloud

  

• Raw & Detected walls & Compressed walls

  

---

Features

• Downsample raw LiDAR data: Uses a PCL VoxelGrid filter to reduce point cloud density by partitioning space into 3D voxels and replacing points in each voxel with their centroid.
• Vertical planar segmentation: Fits plane models to the point cloud using RANSAC, isolating large vertical planes (walls) as inlier points.
• Height filtering: Applies a PCL PassThrough filter on the z-axis to keep points within a specified vertical range (e.g. ground to ceiling).
• Wall compression: Projects detected wall points onto the z=0 plane using PCL’s ProjectInliers (plane) filter, and then applies a fine VoxelGrid downsampling (3 cm leaf size) to produce a 2D representation of the walls.
• Real-time processing: Designed to run online; VoxelGrid downsampling greatly reduces data size, enabling faster RANSAC segmentation and lower computational load.
• ROS-native: Built on ROS Noetic with standard messages (PointCloud2), compatible with existing Velodyne drivers and PCL-ROS tooling.

---

ROS Topics

• Subscribed Topics:

  • /velodyne_points (sensor_msgs/PointCloud2): Raw 3D scan from the Velodyne VLP-16 LiDAR.

• Published Topics:

  • /detected_wall (sensor_msgs/PointCloud2): Point cloud of extracted wall (vertical plane) points.
  • /compressed_wall (sensor_msgs/PointCloud2): Planar-projected, downsampled wall points (all z=0).

---

Pipeline

(1) wall_detect_node.cpp

1. Subscribe to /velodyne_points: Receives raw PointCloud2 messages from the LiDAR.
2. VoxelGrid Filter: Apply a PCL ApproximateVoxelGrid filter to downsample the cloud. A 3D grid of tiny boxes is overlaid on the points and each occupied voxel is replaced by the centroid of its points.
3. ROI (PassThrough) Filter: Use a PCL PassThrough filter on the x, y, z-axis to remove points out of ROI (e.g. keep 0 m < z < 2 m). This focuses the segmentation on plausible walls.
4. Planar Segmentation (RANSAC): Run pcl::SACSegmentation with SACMODEL_PLANE and SAC_RANSAC to find the dominant planar surface in the filtered cloud. Points within a distance threshold of the fitted plane are marked as inliers. In effect, this extracts points belonging to the largest vertical plane (i.e., a wall) in the scene.
5. Publish /detected_wall: The inlier points (the segmented wall) are published as a new PointCloud2. These points represent the detected wall surfaces in the LiDAR view.

(2) wall_compress_node.cpp

1. Subscribe to /detected_wall: Receives the wall point cloud from the first node.
2. Project to Ground Plane: Use PCL’s ProjectInliers filter with a plane model of ax+by+cz+d=0 (with a=b=d=0, c=1) to project all points onto the z=0 (XY) plane. This “flattens” the wall into 2D by setting all z-coordinates to 0.
3. VoxelGrid Downsampling: Apply another ApproximateVoxelGrid filter to the projected points. This coarsely downsampled grid significantly reduces the number of points while preserving the wall outline.
4. Publish /compressed_wall: The result is a sparse 2D representation of the wall, published as PointCloud2. This compressed wall map is useful for real-time navigation or mapping (e.g., as a costmap or occupancy grid).

Each step uses standard PCL/ROS filters (e.g. pcl_ros::ApproximateVoxelGrid, pcl_ros::PassThrough, pcl_ros::SACSegmentation, pcl_ros::ProjectInliers).

---

Dependencies

• ROS Noetic (tested on Ubuntu 20.04) with roscpp and sensor_msgs – for the ROS framework and message types.

• Point Cloud Library (PCL) 1.9+ with pcl_ros and pcl_conversions – for point cloud filtering and segmentation. PCL-ROS is the preferred bridge for 3D point cloud processing in ROS.

• Velodyne drivers (e.g. velodyne_pointcloud, velodyne_driver) – to produce /velodyne_points from the hardware.

---

Installation

Clone the package into your ROS workspace and build it with catkin:

cd ~/catkin_ws/src
git clone https://github.com/Hannibal730/3D-Lidar-wall-detection_ws.git
cd ~/catkin_ws
catkin_make -DCMAKE_BUILD_TYPE=Release
source devel/setup.bash

This will compile the wall_detect_node.cpp and wall_compress_node.cpp into executables. Make sure your Velodyne LiDAR is configured and drivers are running (producing /velodyne_points) before launching this package.

---

Launch Instructions

A convenient launch file wall_detection.launch starts both nodes. For example:

roslaunch wall_detection_pkg wall_detection.launch

This launch file should handle node startup and any necessary topic remappings. Ensure that a running roscore and the Velodyne point cloud publisher are active before running the above command. After launching, you will see /detected_wall and /compressed_wall topics being published.

---

Potential Applications

• Autonomous Vehicles & HD Mapping: Detecting vertical structures (walls, building facades, road barriers) from LiDAR is critical for detailed environment mapping. High-density LiDAR scans with vertical surface information enable the creation of precise 3D urban models and high-definition maps for self-driving navigation.
• Indoor Mobile Robotics: Wall detection aids navigation and SLAM in corridors or rooms (e.g. wall-following robots, indoor mapping). Identifying walls simplifies planning by providing clear boundaries.
• Surveying & Construction: Automated extraction of planar building surfaces from scans can be used for creating as-built building models or checking structural layouts.
• Augmented Reality & Simulation: Fast wall extraction helps reconstruct 3D scenes from LiDAR scans, useful in AR applications or simulated environments where knowing vertical boundaries is important.

---

Questions or Issues

If you have any questions or issues, feel free to ask them in the Issues section.

---

Lisence

This project is released under the Apache License 2.0