Depth camera extrinsics calibration

This package provides tools for calibrating the camera pose and the camera depth offsets by using ROS and OpenCV

Depth offsets calibration (for openni and openni2)

If you need a precise depth information, you have realized that by default the depth given by openni(2) is not perfectly accurate and allows you to specify in its driver z_offset_mm and z_offset_scaling so that:

z = z * z_offset_scaling + z_offset_mm

Currently openni(2) doesn't give any way of calibrating these two values. Lucky you this package does. What happens here is that I use a simple ARMarker and the ROS package ar-track-alvar, I get a set of several poses of the tag and compare the real depth value given by the ARMarker and the depth given by the RGBD sensor (Kinect, Xtion, ...).

Instructions

1 - Calibrate the instrinsics :

http://wiki.ros.org/openni_launch/Tutorials/IntrinsicCalibration

2 - Get the depth_cam_tools package which contains classes for running this code :

https://github.com/kuka-isir/depth_cam_tools

3 - Get ar-track-alvar :

sudo apt-get install ros-$ROS_DISTRO-ar-track-alvar

4 - Create a ARMarker, print it, stick it to a hard object

5 - Run the RGBD sensor with depth registration:

roslaunch openni_launch openni.launch depth_registration:=true

6 - Run the depth offsets calibration :

roslaunch depth_cam_extrinsics_calib calib_depth_offsets.launch

7 - Move the ARMarker in the front of the sensor

8 - Kill the node with Ctrl+C, it will ask you if you want to save the calibration or not to the file specified in the args

Arguments

• kinect_type (string, default: Kinect)

  Can be Kinect or Xtion, no difference so far

• kinect_name (string, default: camera)

• output_file (string, default: $(find kinects_human_tracking)/config/$(arg kinect_name)_depth_offsets.yaml)

  Path to the file you where you want to save the offsets calibration

• nb_points (int, default: 12)

  Number of points to use for calibration

• dz (double, default: 0.05)

  Minimum distance between each points used for calibration

• marker_id (int, default: 0)

  ID of the tag you are using

• marker_size (double, default: 4.0)

  Marker size in cm

• cam_image_topic (string, default: $(arg kinect_name)/rgb/image_mono)

• cam_info_topic (string, default: $(arg kinect_name)/rgb/camera_info)

Sensor pose calibration (checkerboard)

Launch openni (1 or 2) with depth registration.

roslaunch openni_launch openni.launch depth_registration:=true camera:=kinect

Then launch the calibration script (example) :

rosrun kinect_extrinsics_simple simple_chessboard_calib.py kinect /base_link 6 8 0.067 0 0 0
# Arguments are : camera_name base_name chess_width chess_height square_size x0 y0 z0

Note : width is the number of squares in the desired 'x' direction, height in the 'y'. Also, it is preferable to set x0,y0,z0 to zero and publish a static transform from the real base to the upper left corner of the chessboard

Sensor pose calibration (reflexive markers such as Optitrack's or Vicon's)

If you happen to have three reflective markers, a 3D printer and a robot, you can place the 3 markers on the robot and use them to calibrate the camera instead. Feel free to ask for help in the issue's tab

roslaunch depth_cam_extrinsics_calib calib_extrinsics.launch

Arguments

• kinect_type (string, default: Kinect)

  Can be Kinect or Xtion, no difference so far

• kinect_name (string, default: camera)

• base_frame (string, default: /base_link)

• calibration_frame (string, default: /calib_link) Frame attached to the robot that is in the middle of the 3 markers

• output_file (string, default: $(find kinects_human_tracking)/launch/$(arg kinect_name)_extrinsics.launch)

  Path to the file you where you want to save the camera extrinsics calibration