ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question
3

Simple odometry with kinect

asked 2011-11-30 00:34:54 -0500

Chrimo gravatar image

updated 2016-10-24 09:01:54 -0500

ngrennan gravatar image

Hello community,

is it possible to generate any simple odometry informations from kinect sensor ?

Thanks in advance Cheers Christian

edit retag flag offensive close merge delete

4 Answers

Sort by ยป oldest newest most voted
2

answered 2011-11-30 00:52:32 -0500

DimitriProsser gravatar image

Generating odometry information from a camera or depth-sensor like a Kinect is an incredibly difficult problem. Vslam is an attempt to generate visual odometry data from a stereo-camera setup. The Kinect can provide you with the same information as a stereo camera system, in that it publishes an image and a depth map. I believe there have been attempts to use vslam with the Kinect before, but I'm not sure what kind of reliability people have been getting.

Building a 3D obstacle map using SLAM and gmapping can easily be done with the Kinect, however. You can find an example here.

edit flag offensive delete link more

Comments

Building a 3D obstacle map using SLAM and gmapping can be easily done with Kinect instead of a laser scanner, but not without an odometry source. The question was to obtain odometry from laser. I've tried that with Kinect, kinect_to_laser, and laser_scan_matcher. But the results were rather bad
tom gravatar image tom  ( 2011-11-30 01:34:28 -0500 )edit
1

answered 2011-11-30 01:06:17 -0500

phil0stine gravatar image

Yes you can generate odometry estimates from a Kinect, although I wouldn't call it a "simple" process.

There will always be issues with Visual Odometry when there aren't great features, and there are various feature descriptors to choose from that perform differently.

Check rgbdslam for an example of applied visual odom.

edit flag offensive delete link more
1

answered 2011-11-30 01:28:18 -0500

tom gravatar image

updated 2011-11-30 01:57:44 -0500

It's definitely technically complex. But if you're just looking for an implementation, there is a CCNY's experimental stack in which there is a package called kinect_pose_est. Works as presented here. There is a paper to it, see: http://robotics.ccny.cuny.edu/blog/People/Dryanovski , guess it's the number 1 paper. Hope it helps.

Clone the package via:

git clone http://robotics.ccny.cuny.edu/git/ccny-ros-pkg/ccny_experimental.git

Also see this question: http://answers.ros.org/question/502/how-to-use-canonical_scan_matcher-with-kinect

Regards.

edit flag offensive delete link more
0

answered 2011-11-30 18:47:03 -0500

Chrimo gravatar image

Hi,

seems to be a more complex process ;-) I think for the first tries with my new own robot-vehicle too oversized...

What I'd like to try is much more simple: My robot should just move around the rooms and avoid to hit obstacles :-)

What are the minimum requirements for this ?

My hardware consists of the following modules: 1. Motorcontroller based on rosserial_arduino 2. GPSD (useless, but available) 3. Kinect

What do I need more for autonomous moves at unknown environment ?

Thanks for any hints or ideas Cheers Christian

edit flag offensive delete link more

Comments

Hi Chrimo, make it another question please, don't ask more than one question in one post, and if so, do it by editing your original post. Usually the simplest source of odometry information on ground robots are wheel encoders. Also, mark one of the answers best if it answers your (original) question
tom gravatar image tom  ( 2011-11-30 18:57:45 -0500 )edit

Question Tools

2 followers

Stats

Asked: 2011-11-30 00:34:54 -0500

Seen: 3,638 times

Last updated: Nov 30 '11