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

how many reference images used in kinect?

asked 2012-12-03 20:58:50 -0500

this post is marked as community wiki

This post is a wiki. Anyone with karma >75 is welcome to improve it.

  • I want to construct a depth camera like kinect.
  • One referece image is sanped at a distance of 100cm, and servel test images are snapped at different distances. I found if the test image with a distance not similar to 100cm, the disparity can not be calculated correctly. Whether I miss something when I calculate the disparity?

  • Is there someone konw how many reference images kinect used?

edit retag flag offensive close merge delete

3 Answers

Sort by » oldest newest most voted

answered 2012-12-04 16:18:18 -0500

This paper provides a good overview of the Kinect measurement model:

Accuracy and Resolution of Kinect Depth Data for Indoor Mapping Applications

In the paper, the authors refer to a parameter Z_0, or the distance to the reference plane. I assume this is what you're asking about. They calibrate for the depth readings, so if you go through the equations, you can find the value of Z_0.

Now in the model the authors describe, they only use a single Z_0. However, it is entirely possible that the Kinect uses multiple and switches dynamically, but that could be hidden from the user.

edit flag offensive delete link more

answered 2012-12-03 21:58:51 -0500

Ben_S gravatar image

The Kinect does not use reference images of any kind, its more of an active stereo vision approach. There is a known infrared-pattern projected by a laser (this serves as the "right eye") and an camera with an ir-filter as the left eye. This article gives a quick overview. Or with a little marketing in this document.

edit flag offensive delete link more

answered 2012-12-04 15:12:27 -0500

this post is marked as community wiki

This post is a wiki. Anyone with karma >75 is welcome to improve it.

Thans Ben_S But, if the kinect uses the stereo vision approach, it needs to know the corresponding position in right image of each point in left image, used for calculating the disparity. But the IR-projector only provides a binary pattern, so it needs to snap a image as reference at a known depth.

edit flag offensive delete link more


Hi beigua - please consider commenting when reply to an answer, instead of posting new answers.

Ivan Dryanovski gravatar image Ivan Dryanovski  ( 2012-12-04 16:19:00 -0500 )edit

thanks Dryanovski, your answer and comment is very helpful.

beigua gravatar image beigua  ( 2012-12-04 19:36:54 -0500 )edit

Question Tools

1 follower


Asked: 2012-12-03 20:58:50 -0500

Seen: 641 times

Last updated: Dec 04 '12