s-arora1987 / sanet_onionsorting

sanet_onionsorting

Author: Prasanth Sengadu Suresh.

Owned By: THINC Lab, Department of Computer Science, University of Georgia.

This package uses Kinect V2 and YOLO to detect and obtain 3D coordinates of objects in the real world.

The following instructions are written for Ubuntu 18.04, ROS Melodic.

If you need to install the packages for Kinect V2 with Ubuntu, check out this link.

The following are the steps to be followed to get this package working:

Assuming you have a working version of Ubuntu (This package has been built and tested on Ubuntu 18.04)

1.) Install ROS (This package was built on ROS Melodic, for other versions you might need to make appropriate changes)

ROS Melodic

Catkin Workspace

2.) Clone this package and catkin_make within catkin_ws dir.

Note: There are some files with absolute paths of directories/files used within them. If you get a FileNotFound/DirectoryNotFound error, please open the corresponding file and check. I'm in the process of converting them to relative paths.

3.) For the location of the camera used in this package, check this link if you're using a real kinect or this link if you're using this kinect_v2 package with Gazebo.

Note: If you plan to use your kinect elsewhere, you have to change the coordinates accordingly in the static_transform publisher and the kinect_v2_standalong_physical.urdf.xacro file.

4.) For the real kinect:

  roslaunch kinect2_bridge kinect2_bridge.launch
  
  roslaunch kinect2_bridge rviz_tf.launch
  
  rosrun sanet_onionsorting yolo_service.py real
  
  rosrun sanet_onionsorting rgbd_imgpoint_to_tf.py real

For Gazebo kinect: After launching your required simulation setup on Gazebo:

    roslaunch kinect_v2 kinect_v2_full.launch
    
    rosrun sanet_onionsorting yolo_service.py gazebo

    rosrun sanet_onionsorting rgbd_imgpoint_to_tf.py gazebo

5.) Now if you listen to /object_location topic, you get the 3D world coordinates of objects recognized by YOLO.