Clustering for 3D Pointclouds
pointcloud_segmentation is a package that provides some tools to segment 3D pointclouds. A detailed description about c++ node classes and nodelets are given below.
Here is a brief description on how to use the nodelets and node classes.
This node will take a message of type sensor_msg::PointCloud2 and pass a voxel grid filter and a volume filter. Usefull tool for stereo pointclouds.
<!-- FILTER XB3 POINT CLOUD -->
<node name="cloud_filtering" pkg="pointcloud_segmentation" type="filtering_nodelet" >
<param name="pointcloud_input" type="string" value="/snr/scam/wide/points2" />
<param name="pointcloud_output" type="string" value="/snr/scam/wide/filtered/points2" />
<param name="frame_id" type="string" value="/atc/vehicle/center_bumper" />
<param name="voxel_size" type="double" value="0.05" /> Voxel resolution
<param name="min_x" type="double" value="0.0" /> Min for x value relatively to /atc/vehicle/center_bumper
<param name="max_x" type="double" value="15.0" /> Max for x value relatively to /atc/vehicle/center_bumper
<param name="min_y" type="double" value="-8.0" /> Min for y value relatively to /atc/vehicle/center_bumper
<param name="max_y" type="double" value="8.0" /> Max for y value relatively to /atc/vehicle/center_bumper
<param name="min_z" type="double" value="-0.5" /> Min for z value relatively to /atc/vehicle/center_bumper
<param name="max_z" type="double" value="5.0" /> Max for z value relatively to /atc/vehicle/center_bumper
</node>
Actualy this nodelet is similar to filtering_nodelet but it accepts only laser_scan instead of pointclouds
<h2> <a name="laser_to_pc_nodelet"> laser_to_pc_nodelet </a> </h2> <node name="laser_to_cloud" pkg="pointcloud_segmentation" type="laser_to_pc_nodelet"> <param name="laser_input" type="string" value="/snr/las/2/scan" /> <param name="pointcloud_output" type="string" value="/snr/las/2/filtered/points2" /> <param name="frame_id" type="string" value="/atc/vehicle/center_bumper" /> <param name="min_x" type="double" value="0.0" /> <param name="max_x" type="double" value="15.0" /> <param name="min_y" type="double" value="-8.0" /> <param name="max_y" type="double" value="8.0" /> <param name="min_z" type="double" value="-0.5" /> <param name="max_z" type="double" value="5.0" /> </node>
This nodelet requires the use of package plane_model_road_segmentation to make a RANSAC model for road plane. It will accept a message of type sensor_msg::PointCloud2 and return two messages: one with road pointcloud and other with the rest of the points.
<!-- PLANE MODEL ROAD SEGMENTATION -->
<node name="plane_model_road_extraction" pkg="pointcloud_segmentation" type="plane_road_extraction_nodelet">
<param name="pointcloud_input" type="string" value="/snr/scam/wide/filtered/points2" />
<param name="pointcloud_road" type="string" value="/snr/scam/wide/road/points2" />
<param name="pointcloud_clusters" type="string" value="/snr/scam/wide/clusters/points2" />
<param name="z_min" type="double" value="-0.5" /> Minimum value for road extraction value
<param name="z_max" type="double" value="0.15" /> Maximum value for road extraction value
<param name="threshold" type="double" value="0.15" /> Threshold limit for plane_model_road_segmentation class. see /lar3/utils/plane_model_road_segmentation
</node>
This nodelet will accept a pointcloud for euclidean segmentation. It will assume that the input data is already pre-processed and there is no contact between obstacles so you can set properly your parameters.
<!-- EUCLIDEAN CLUSTERING --> <node name="euclidean_cluster" pkg="pointcloud_segmentation" type="euclidean_cluster_nodelet"> <param name="pointcloud_input" type="string" value="/snr/scam/wide/clusters/points2" /> <param name="cluster_markers" type="string" value="/environment/clusters/marker" /> <param name="cluster_markers_text" type="string" value="/environment/clusters/info" /> <param name="cluster_tolerance" type="double" value="0.5" /> <param name="min_cluster_size" type="double" value="5" /> <param name="max_cluster_size" type="double" value="10000" /> </node>
This nodelet have as input two diferent pointclouds: one is from a laserscan or similar to make euclidean cluster segmentation, and then it will accept a 3D pointcloud for neighboor search.
<!-- REGION GROWING --> <node name="region_growing" pkg="pointcloud_segmentation" type="region_growing_nodelet"> <param name="pointcloud_input" type="string" value="/snr/scam/wide/clusters/points2" /> <param name="laser_input" type="string" value="/snr/las/2/filtered/points2" /> <param name="cluster_markers" type="string" value="/environment/clusters/markers" /> <param name="cluster_markers_text" type="string" value="/environment/clusters/info" /> <param name="min_cluster_size" type="double" value="2" /> <param name="max_cluster_size" type="double" value="1000" /> <param name="clustering_tolerance" type="double" value="1.0" /> <param name="radius" type="double" value="0.5" /> </node>
If you need to configure your own nodelets for other applications you can use the class nodes as well. Each nodelet uses a c++ class and you can use them as your own will :)