doctorwk007/point-cloud-clusters

A catkin workspace in ROS which uses DBSCAN to identify which points in a point cloud belong to the same object.

point-cloud-clusters

I apply a clustering technique called DBSCAN to identify which points in a point cloud belong to the same object.

• This is the second perception exercise from Udacity's RoboND.

• This builds upon my solution for the first perception exercise where I apply techniques to separate our objects of interest.

Important Files

• https://github.com/mithi/point-cloud-clusters/blob/master/src/sensor_stick/scripts/clustering.py
• https://github.com/mithi/point-cloud-clusters/blob/master/src/sensor_stick/scripts/filtering_helper.py
• https://github.com/mithi/point-cloud-clusters/blob/master/src/sensor_stick/scripts/pcl_helper.py

Related links

You can learn more about PCL here. You can learn more about DBSCAN in the following links:

• https://www.naftaliharris.com/blog/visualizing-dbscan-clustering/
• http://scikit-learn.org/stable/auto_examples/cluster/plot_dbscan.html

Dependencies

• You need an Ubuntu 16.04.2 with ROS full-desktop-version which includes RViz and Gazebo
• You must clone the repository, go inside the directory and install the dependencies:

$ rosdep install --from-paths src --ignore-src --rosdistro=kinetic -y
$ catkin_make

• Add the following to your .bashrc file:

export GAZEBO_MODEL_PATH=~/catkin_ws/src/sensor_stick/models
source ~/catkin_ws/devel/setup.bash

How to Run

• On one terminal run $ roslaunch sensor_stick robot_spawn.launch
• On another terminal go inside /src/sensor_stick/scripts/ folder in this repository
• Then run $ python clustering.py
• RViz should run, select the /pcl_cluster from the Topics dropdown