Repository files navigation

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

• 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)

For new projects:

mkdir catkin_ws/srccd catkin_ws/srcgit clone https://github.com/paul-shuvo/planar_pose.gitcd planar_posepip install -r requirements.txtcd ../../..catkin_make

For existing project:

cd 'your_project_path'/srcgit clone https://github.com/paul-shuvo/planar_pose.gitcd planar_posepip install -r requirements.txtcd ../../..catkin_make

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.bashrosrun planar_pose object_detection.py

Then, to run the planar pose estimation module:

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

• /detected_objectCoordinates 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_infoPose (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_arrayPoseArray of the detected objects.

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

The images of the objects' planes should be put inside thesrc/objects folder. The path could be changed in theconfig.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

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},    }

• Do more tests