SAMXL_pointcloud_area_detection

Overview

This repository performs volume loss calculation between two point clouds: from before grinding and from after grinding, scanned by laser line scanner. The node works in conjuction with data_gathering repositories from [email protected]:Luka140/data_gathering.git.

Installation

Dependencies (INCOMPLETE)

• ROS2 Humble
• pyads
• open3d
• sklearn
• scipy
• concave_hull

pip install pyads
pip install open3d==0.18.0
pip install numpy==1.24.0
pip install scikit-learn==1.5.2
pip install scipy==1.8.0
pip install git+https://github.com/panin-anan/concave_hull.git

sudo apt-get update
sudo apt install ros-humble-rosbag2-storage-mcap

Building

Note: This README shows the bare minimum to use the manual_publish_and_process_pcl.launch.py launch file. To use this repository its full capability, navigate to [email protected]:Luka140/data_gathering.git and follow its README install all necessary repositories

To build from source, clone the latest version from this repository into your workspace along with the following required repositories:

• data_gathering_msgs : Server for the Service RequestPCLVolumeDiff

git clone [email protected]:Luka140/data_gathering_msgs.git
git clone [email protected]:panin-anan/pcl_processing_ros2.git

and compile the packages: To build, after cloning the package:

colcon build --symlink-install
install/setup.bash

Launch files

This repository contains two launch files. Both require similar launch parameters which is explained in the Nodes section.

scan_and_process_pcl.launch.py

only launch pcl_processing node which waits for a ROS2 service request to the service RequestPCLVolumeDiff before performing volume detection using the request message.

manual_publish_and_process.launch.py

Launch three nodes: pcl_processing, pcl_publish_manual, and rviz. The pcl_publish_manual node waits for manual keyboard input to publish pointcloud onto the Service Request and, when two pointclouds are manually published, initiate the volume detection by pcl_processing node.

Nodes

pcl_processing

A node that automates the volume loss calculation process. The node is a service server which waits for the Service Request, RequestPCLVolumeDiff, before the node computes the volume loss from the latest two point clouds stored in a service client. This service client is created either in test_coordinator node from data_gathering package or in pcl_publush_manual node for manual volume loss calculation, according to the launch file used.

The volume loss is calculated from the point cloud difference between two point clouds, clouds from before and after grinding, using the following process: Note: For now, the algorithm is only applicable to flat plate

1. Uses RANSAC to segment the two set of point clouds and determine the plane parameters of the plate, along with DBSCAN clustering to filter out points unrelated to the sample plate.
2. Project points onto the same plane and uses KDTree to search for nearest neighbours for difference in point-point distance: Any points which are a certain laserline_threshold or feedaxis_threshold distance away from the post grind points are points that have disappeared from compared to the pre-grind point cloud.
3. DBSCAN clustering again to filter out outliers
4. Calculate volume lost from area * plate_thickness in which the area is calculated using concave hull algorithm to find the area from the latest clustered 2D point cloud.

Parameters:

• dist_threshold: out of plane distance to include points on the plane of the sample plate from RANSAC
• cluster_neighbor: minimum number of points for a cluster in DBSCAN clustering
• plate_thickness: sample plate thickness to be used to calculate volume loss
• plane_error_allowance: allowable errors in plane angle between two point clouds before throwing error.
• clusterscan_eps: minimum distance for grouping cluster in point difference clustering algorithm
• laserline_threshold: minimum distance threshold for detecting point cloud difference in the laserline axis (finer resolution axis)
• feedaxis_threshold: minimum distance threshold for detecting point cloud difference in the feedaxis axis (coarser resolution axis)
• concave_resolution: resolution for concave hull: high = tight-knitted area, low = convex hull
• filter_down_size: voxel down size of scanned data for clustering algorithm to speed up clustering process
• belt_width_threshold: threshold for checking a valid pointcloud difference detection based on grinding belt standard width.

Topics:

• /scanner/volume Float32Stamped: Publishes the calculated volume lost during the grinding process.
• /grinded_cloud PointCloud2: Publishes the point cloud of the grinded surface after volume loss calculation.
• /concave_hull_lines Marker: Publishes concave hull lines as markers for visualization of the calculated hull from the point cloud.

Services:

• calculate_volume_lost RequestPCLVolumeDiff: A service that calculates the volume lost between two point clouds, taking initial and final point clouds as inputs, and returning the volume difference and the processed point cloud.

pcl_publish_manual

A node that can be used to manually publish saved point cloud to pcl_processing node from selected point cloud files(.ply), in order to manually evaluate tested data/point clouds. use Key P to initiate UI and load desired pairs of point cloud for evaluation. The point clouds are also visualize in rviz.

Parameters:

• dist_threshold: out of plane distance to include points on the plane of the sample plate from RANSAC
• cluster_neighbor: minimum number of points for a cluster in DBSCAN clustering
• plate_thickness: sample plate thickness to be used to calculate volume loss
• plane_error_allowance: allowable errors in plane angle between two point clouds before throwing error.
• clusterscan_eps: minimum distance for grouping cluster in point difference clustering algorithm
• laserline_threshold: minimum distance threshold for detecting point cloud difference in the laserline axis (finer resolution axis)
• feedaxis_threshold: minimum distance threshold for detecting point cloud difference in the feedaxis axis (coarser resolution axis)
• concave_resolution: resolution for concave hull: high = tight-knitted area, low = convex hull
• filter_down_size: voxel down size of scanned data for clustering algorithm to speed up clustering process
• belt_width_threshold: threshold for checking a valid pointcloud difference detection based on grinding belt standard width.

Topics:

• /combined_cloud PointCloud2: Publishes the point cloud to rviz when key P is pressed and point clouds loaded

Services:

• calculate_volume_lost RequestPCLVolumeDiff: A service that calculates the volume lost between two point clouds, taking initial and final point clouds as inputs, and returning the volume difference and the processed point cloud.