Segmenter

ROS wrapper for RGBD segmentation.

Description

This package is a ROS wrapper that performs segmentation based on RGBD segmentation by Richtsfeld et al. This package can be used in junction with existing grasping system, i.e., fetch_grasp_suggestion. This package also includes functionality to train Support Vector Machine model using libsvm.

Installation

This package was tested on Ubuntu 14.04

1. Install ROS-indigo. Installation instructions can be found here. Full Desktop installation is recommended.
2. Install the package by cloning the segmenter_jordlee repository into your catkin workspace and building it as follows:

   cd (your catkin workspace)/src
   git clone https://github.com/fetchrobotics/sandbox/segmenter_jordlee.git
   cd ..
   catkin_make install

Usage

segmenter

1. change model_path directory and string point_cloud_topic into your local directory in 'Segmenter::Segmenter' in src/segmenter_jordlee.cpp

2.  cd (your catkin workspace)
    source devel/setup.bash 
    export ROS_MASTER_URI=http://<robot_name_or_ip>:11311

3. run rosnode

   rosrun segmenter_jordlee segmenter_jordlee

4. open a new terminal. call rosservice. (repeat step 1, 2 before rosservice call)

   rosservice call /segment_object "{}" > /dev/null

5. Visualization in Rviz

   rosrun rviz rviz

   change frame to base_link. Pointcloud before segmenting will be under topic sensor_msgs/PointCloud2/pointtestinginput.
   Segmented objects will be under topic visualization_msgs::MarkerArray/markers_jordlee

svm_model_creator

Currently, the segmenter uses 'PP-Trainingsset.txt.scaled.model' as a default SVM model. This is trained by Object Segmentation Database. User can change this by using their own dataset. mOSD trained data set is provided as an example in the src/svm_model_creator.cpp

1. prepare own object data set

2.     cd (your catkin workspace)/devel/lib/segmenter_jordlee
       ./svm_model creator -f ~/path_to_your_objectdatasets/datasets%1d.pcd -idx 0(number of your current train set)

3. press F10 once TomGine window pops up.

4. Go to src/svm_model_creator.cpp

5. go to 'Segmenter::annotator' function.

   • anno[number of surfacepatch of interest].pushback(numbering of surface patch next to it that is within same object)
   • repeat this for all surfaces except the floor and surfaces which are extremely small
   • move to next surface This annotation handles double counting so you do not need to count it twice (i.e., anno[1].pushback(2), no need to anno[2].pushback(1))

6. when you finished the annotation for all the datasets,

   ./svm_model creator -f ~/path_to_your_objectdatasets/datasets%1d.pcd -idx 0 10(total count of your training data set)

   this will generate 'model.txt' in '/devel/lib/segmenter_jordlee'

7. move this model.txt to model folder in your src/segmenter_jordlee/model

8. change of variable name is segmenter_jordlee.cpp for using this model.

Primary ROS Nodes

• Subscribers
  • /head_camera/depth_registered/points(pcl/PointCloud<pcl/PointXYZRGB>)
    Point cloud stream used for updating the scene for segmentation. The point cloud subscribed to can be changed by setting the string point_cloud_topic in constructor of 'Segmenter::Segmenter'.
• Publishers
  • /segmented_objects(rail_manipulation_msgs/SegmentedObjectList)
    segmented object list
  • /pointstestinginput(pcl/PointCloud<pcl/PointXYZRGB>)
    pointcloud before segmenting. Used for visualization in rviz.
  • /segmented_markers(visualization_msgs/MarkerArray)
    visualization markers in rviz
• Service Clients `/segment_object(rail_manipulation_msgs/SegmentedObjectList)
  segmented object list. This has a same type with the segmented object list in the publisher. So the same message types are both published and saved in the service response. Users can remove one of these for their own need.

Connecting with grasping system

This version currently works with fetch_grasp_suggestion, which subscribes rail_segmentation/SegmentedObjectList. Users may change the type of the message for their grasping system. Please refer to connecting with alternative segmenter for detailed instruction.

Authors

ROS Wrapper

Ung Hee(Jordan) Lee
Mechanical Engineering
University of Michigan
unghee@umich.edu

RGB-D Segmentation

Dipl.-Ing. Andreas Richtsfeld
Automation and Control Institute (ACIN)
Vienna University of Technology
Gusshausstraße 25-29 1040 Vienna
ari(at)acin.tuwien.ac.at

Citation

1. Richtsfeld A., Mörwald T., Prankl J., Zillich M. and Vincze M. - Segmentation of Unknown Objects in Indoor Environments. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2012 (pdf)
2. Chih-Chung Chang and Chih-Jen Lin, LIBSVM : a library for support vector machines. ACM Transactions on Intelligent Systems and Technology, 2:27:1--27:27, 2011. Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm
3. strands-project v4r library
4. Modified Object Segmented Datasets

Developer note

The speed of running the package decreased after including and building from source file on_nurbs opennurbs from strands-project v4r library instead of using pcl 1.7.2. For now, easy solution would be install the pcl 1.7.2. and connect the header files to the pcl instead of this stand alone on_nurbs opennurbs.