paul-shuvo / ROS-Pose

A ROS (Robotic Operating System) package for simple object detection and planar pose estimation for objects that requires only an image of the plane of the object facilitating quick prototyping of Robotics applications.

Planar Pose Estimation

A ROS package for a simple object detection and planar pose estimation method for textured objects.

• Planar Pose Estimation
  • Dependencies
  • Install
  • Run
  • Published Topics
  • Demo
  • Citation
  • TODO

Dependencies

• ROS Noetic/Melodic/Kinetic
• Python 2.7, 3.3+ for ROS Noetic (should support Python 3 for other ROS Distros as well)
• OpenCV (Tested on opencv 4, should work with other opencv versions as well)

Install

For new projects:

mkdir catkin_ws/src
cd catkin_ws/src
git clone https://github.com/paul-shuvo/planar_pose.git
cd planar_pose
pip install -r requirements.txt
cd ../../..
catkin_make

For existing project:

cd 'your_project_path'/src
git clone https://github.com/paul-shuvo/planar_pose.git
cd planar_pose
pip install -r requirements.txt
cd ../../..
catkin_make

Run

This package contains a object detection module and a planar pose estimation module. Planar pose estimation module depends on the object detection module.

To run the object detection module:

cd 'your_project_path`
source devel/setup.bash
rosrun planar_pose object_detection.py

Then, to run the planar pose estimation module:

cd 'your_project_path`
source devel/setup.bash
rosrun planar_pose planar_pose_estimation.py

Published Topics

• /detected_object Coordinates of the boundary of the detected objects in terms of a json string.

  ---
  "data: {\"cheezeit\": [[369, 252], [367, 344], [423, 339], [429, 251]], \"book-1\": [[518, 234],\
  \ [525, 225], [513, 230], [590, 222]]}"
  ---

• /object_pose_info Pose (Quaternion) info of the detected objects in terms of a json string.

  ---
  data: "{\"cheezeit\": {\"position\": [-0.21074476838111877, 0.17126992344856262, 1.377000093460083],\
  \ \"orientation\": [-0.7785678371108322, -0.03650856877418329, 0.6243951877330678,\
  \ 0.05128252098335612]}, \"book-1\": {\"position\": [0.10074007511138916, 0.07484704256057739,\
  \ 0.8750000596046448], \"orientation\": [-0.99671235271185, -0.024004214965333632,\
  \ -0.05656042965190621, 0.052812890595937564]}}"
  ---

• /object_pose_array PoseArray of the detected objects.

  ---
  header: 
      seq: 9
      stamp: 
          secs: 1612235248
          nsecs: 180106401
      frame_id: "kinect2_rgb_optical_frame"
  poses: 
    - 
      position: 
          x: -0.3775361180305481
          y: 0.18009251356124878
          z: 1.2940000295639038
      orientation: 
          x: -0.8238780569888
          y: -0.005098909875330737
          z: 0.5370874941853346
          w: 0.1809308484481323
      - 
      position: 
          x: 0.04039544239640236
          y: 0.06119309365749359
          z: 0.8650000691413879
      orientation: 
          x: -0.997179603918241
          y: -0.036229147670746836
          z: -0.0050051373858676085
          w: 0.06553804229804013
  ---

Demo

The images of the objects' planes should be put inside the src/objects folder. The path could be changed in the config.py file. For the demo below, the following image of a "Cheezit" box was used.

(a), (b), (c) are recovered poses from robot’s camera and (d), (e), (f) are corresponding poses visualized in RViz

Citation

Please cite these papers in your publications if it helps your research:

@inproceedings{paul2020object,
      title={Object Detection and Pose Estimation from RGB and Depth Data for Real-time, Adaptive Robotic Grasping}, 
      author={S. K. Paul and M. T. Chowdhury and M. Nicolescu and M. Nicolescu},
      booktitle={Proceedings of the International Conference on Image Processing, Computer Vision, and Pattern Recognition (IPCV)}
      year={2020},    
}

TODO

• Do more tests