Kinect2 Calibration

[mwswartwout]

Have taken two groups of calibration images.

1st image set

Appears to not have good depth data for calibration. Occasionally will segfault when calibrating with this data set, appears similar to this issue: https://github.com/code-iai/iai_kinect2/issues/66. Initially was able to use the kinect2_viewer with these files, but now depth registration returns some enqueueError. Not planning on investigating this further.

2nd image set

Much better. Calibration works for objects that in the background, but is bad for objects in the foreground.

Both image sets combined

Very bad calibration data.

To Do:

• [x] Throw out first data set
• [x] Take further calibration images where object rotates on axis perpendicular to camera surface (requires remounting calibration image to tripod)
• [x] Take more images with calibration pattern closer to Kinect2
• [x] Test the two datasets independently and then combine to see if results improve
• [x] Research to find example point cloud images for Kinect2 to get point of comparison for what our data should look like

  2nd Dataset Calibration Image

Both Datasets Calibration Image

Uncalibrated Image

[wsnewman]

Matt, normally, camera calibration uses a checkerboard that nearly fills the scene. You might try doing the calibration closer, and also possibly using a larger target.

We care the most about calibration in our "sweet spot" (about arm's length), so I would encourage you to take most of your calibration data in that range.

On Wed, Sep 23, 2015 at 1:00 PM, Matthew Swartwout <notifications@github.com

wrote:

Have taken two groups of calibration images. 1st image set

Appears to not have good depth data for calibration. Occasionally will segfault when calibrating with this data set, appears similar to this issue: code-iai/iai_kinect2#66 https://github.com/code-iai/iai_kinect2/issues/66. Initially was able to use the kinect2_viewer with these files, but now depth registration returns some enqueueError. Not planning on investigating this further. 2nd image set

Much better. Calibration works for objects that in the background, but is bad for objects in the foreground. Both image sets combined

Very bad calibration data. To Do:

• Throw out first data set
• Take further calibration images where object rotates on axis perpendicular to camera surface (requires remounting calibration image to tripod)
• Take more images with calibration pattern closer to Kinect2
• Test the two datasets independently and then combine to see if results improve
• Research to find example point cloud images for Kinect2 to get point of comparison for what our data should look like

[image: calibratedwith2nddata] https://cloud.githubusercontent.com/assets/1334791/10052097/0eefbac4-61f3-11e5-9368-3359dc545b60.png [image: calibratedwithbothdatasets] https://cloud.githubusercontent.com/assets/1334791/10052098/0ef2406e-61f3-11e5-9b0d-543092f6e0a8.png [image: uncalibrated] https://cloud.githubusercontent.com/assets/1334791/10052096/0eed09be-61f3-11e5-8148-84b6f99f0aa3.png

— Reply to this email directly or view it on GitHub https://github.com/cwru-robotics/baxter-model-scan/issues/2.

Wyatt Newman, Ph.D., P.E. Professor, EECS Dept Case Western Reserve University Cleveland, OH 44106 216-368-6432 wyatt.newman@case.edu

[lucbettaieb]

For reference, this is what we were trying to follow: https://github.com/code-iai/iai_kinect2/tree/master/kinect2_calibration#calibrating-the-kinect-one

[mwswartwout]

Intrinsic calibration completed and pull requested to repo. Extrinsic calibration still in progress.

[lucbettaieb]

@mwswartwout Can you update your ticket with what you've been doing, please?

[mwswartwout]

Static tf publisher is working. Have initial "magic numbers" that get calibration close but not quite there. Don't have a method other than guess & check currently. Goal is to have a transform that is good enough by Thursday so that work can begin on more accurate calibration method.

[mwswartwout]

Calibration stalled until Kinect2 Mount is completed.

Going to investigate this: http://wiki.ros.org/openni/Contests/ROS%203D/Automatic%20Calibration%20of%20Extrinsic%20Parameters

[mwswartwout]

Looking at this too: http://wiki.ros.org/industrial_extrinsic_cal

[lucbettaieb]

@mwswartwout Check this out, as well! https://github.com/mikeferguson/robot_calibration

[mwswartwout]

Looking at this too http://wiki.ros.org/calibration

[lucbettaieb]

@mwswartwout Please update your ticket with your progress.

[mwswartwout]

https://github.com/osrf/baxter_demos

This is a Baxter-specific extrinsic calibration program using AR tags. Am investigating and trying to get simulation semi-working in Gazebo.

[lucbettaieb]

Does it work???

[mwswartwout]

Lol no, I just figured I was never going to finish it.