What packages am I going to use to work with Kinect v2 and object recognition

asked 2017-06-15 08:02:54 -0500

roel958 gravatar image

HI there. We are currently working with a group of students on a project. We have a custom build Robot gripper arm, which we should be able to move and pickup objects.

The project exists with the following:

-Intel NUC -- Linux 16.04 LTS -- ROS Kinetic -Kinect v2 (one) -Custom build gripper arm to pick up objects

We are striving to the following goal:

1) Kinect publishes live camera feed 2) Apply object recognition on live feed 3) Recognise object and determine which one to grab 4) Apply movement to gripper to grab the object

As far as my knowledge right now goes we do need to use the following packages:

1) Publish using freenect_launch and libfreenect 2) Using pcl_recognition or Object Recognition Kitchen (ORK) for 3D point cloud processing. Which is better? 3) Using custom cpp or python scripts (?) 4) Same as 3

Things that worked: -Getting all the 4 feeds from the Kinect using libfreenect Protonect

Things that didn't work: -Launch the freenect to publish the Kinect's topic, it says waiting for device to connect -Using iai_kinect2 as a bridge, couldn't install it because of the opencv3 being installed with ROS Kinetic

As i see the following https://github.com/TuZZiX/pcl_recogni... a package for object recognition, if you read the readme. It talks about either pcl_recognition and using ORK. I am a bit confused, can you either pick one or do you use both?

We experimented a lot with installing different packages and trying to find the working solutions. Our question is, are we on the right way? Are we forgetting an important package?

Also I see a lot about opencv being used in different situations, is this something we should use?

Thank you


edit retag flag offensive close merge delete