This package allows for the calibration of two rigidly-linked reference frames given a series of transformations. You can see this package in action in our ROS contest entry: openni/Contests/ROS 3D/Automatic Calibration of Extrinsic Parameters.

There are two programs in this package: tf_logger, a node that dumps the transformations regularly and optimizer, a stand-alone program that computes the transformation between the frames.

For installation instructions, see the ethzasl_icp_mapping stack page.

Program: Logging

tf_logger will record the changes of two frames published in tf each with respect to its own reference and dump it as text into an output file.

rosrun ethzasl_extrinsic_calibration tf_logger


~/baseLinkFrame (string, default: "/base_link")

  • First reference frame to log.

~/odomFrame (string, default: "/odom")

  • Parent frame for the first reference frame.

~/kinectFrame (string, default: "/openni_rgb_optical_frame")

  • Second reference frame to log.

~/worldFrame (string, default: "/world")

  • Parent frame for the second reference frame.

~/outputFileName (string, default: "output.txt")

Program: Calibration

optimizer will use an evolutionary strategy to compute the transformation from the first reference frame to the second, reading these frames from a text file.

rosrun ethzasl_extrinsic_calibration optimizer <dtf_file> [inlier_ratio] [restart_count] [gen_count]


<dtf_file> (filename)

  • Name of output file from tf_logger or from modular_cloud_matcher_bag.

[inlier_ratio] (double, default: 0.8)

  • Ratio of inlier to keep when computing the error. The pair of transforms with the largest errors will be discarded. This allows to find a good-quality transform even if the tracker used to produce the transforms sometimes failed.

[restart_count] (positive integer, default: 1)

  • Number of random restarts of the evolution strategy. Setting this value to more than 1 will run the optimization multiple times starting from random values and will give you the average result and an estimate of its variance.

[gen_count] (positive integer, default: 64)

  • Number of generation for the evolution strategy. Increasing this setting will reduce the variance between random restarts, but will take more time.


optimizer prints both a line that can be added to a launch file to provide the computed transformation in the tf tree and the command to publish this transformation directly from the command line:

Optimization completed, code to COPY-PASTE in to use the transformation:

<node pkg="tf" type="static_transform_publisher" name="base_link_to_kinect"
args=0.182129 0.00632656 0 -0.491826 0.498884 -0.499452 0.509676 /base_link /kinect 100"/>


rosrun tf static_transform_publisher 0.182129 0.00632656 0 -0.491826 0.498884 -0.499452 0.509676 /base_link /kinect 100

Also, if random restarts are used, it gives an estimate of the variance.

File Format

The text file between tf_logger and optimizer consists of a sequence of lines, each line providing a pair of transforms. This, the format for one line is the following:

tr1_trans_x tr1_trans_y tr1_trans_z tr1_quat_x tr1_quat_y tr1_quat_z tr1_quat_w tr2_trans_x tr2_trans_y tr2_trans_z tr2_quat_x tr2_quat_y tr2_quat_z tr2_quat_w

Known issues

On x86-32, sometimes the optimizer segfaults. Either retry or use an amd64 kernel.

Report a Bug

Please report bugs and request features using the github page.

Wiki: ethzasl_extrinsic_calibration (last edited 2012-12-18 21:17:36 by StephaneMagnenat)