Skip to content

Latest commit

 

History

History
139 lines (99 loc) · 5.05 KB

calibration.md

File metadata and controls

139 lines (99 loc) · 5.05 KB
Raw

Calibration

All Core Research Development Kits come with factory intrinsic and extrinsic calibration. We provide this calibration on our servers in several file formats. The driver comes with a command to easily obtain the calibration files belonging to your sensor.

Calibration model

Camera intrinsics

The individual cameras are calibrated using a pinhole projection model. The distortion of the lens is modelled with an equidistant model with five terms. See the reference below for more information on the equidistant model.

ℹ️ Info: The pinhole+equidistant model is not the same as the plumb_bob model that is commonly used in ROS. Because of our wide-angle lenses the equidistant model is better suited.

J. Kannala and S. Brandt (2006). A Generic Camera Model and Calibration Method for Conventional, Wide-Angle, and Fish-Eye Lenses, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28, no. 8, pp. 1335-1340 (pdf link)

Extrinsics

For each camera we provide the 6-DOF transformation with respect to the IMU.

File formats

At the moment we produce two file formats 7s_sensors.yaml and kalibr_calib.yaml. The main calibration file is 7s_sensors.yaml. The others are generated to make our demos easy to use.

7s_sensors.yaml

All calibrated parameters are available in this file. An example can be found here.

For each camera there is a yaml node under ncameras/cameras, like the one below.

- T_B_C: 
    rows: 4
    cols: 4
    data: [
    -0.0089889092, 0.0097254569, 0.9999123037, 0.0520086092,
    -0.9999261367, -0.0082678741, -0.0089086176, 0.0488993112,
    0.0081805087, -0.9999185256, 0.0097990577, -0.011757515,
    0.0, 0.0, 0.0, 1.0]
  camera:
    distortion:
      parameters: 
        rows: 4
        cols: 1
        data: [-0.0416026702, 0.0022689289, -0.0027567794, 0.000401603]
      type: equidistant
    id: 839e02677a1f4b5c91466b2bc5fdc33e
    image_height: 1080
    image_width: 1440
    intrinsics: 
      rows: 4
      cols: 1
      data: [701.4165958679, 701.480279171, 668.2392112416, 517.9783218077]
    label: /alphasense_driver_ros/cam0
    line-delay-nanoseconds: 0
    type: pinhole

The label node identifies to which camera the calibration belongs. In this example the calibration is for camera 0.

  label: /alphasense_driver_ros/cam0

The parameters for the intrinsic calibration are encoded in the camera/intrinsics and camera/distortion field for the projection model and distortion model respectively.

Projection model

The four parameters for the pinhole projection model are specified in the order fu, fv, cu, cv. Where the pair fu fv is the focal length in u and v directions on the image plane, and the pair cu cv is the location of the center of projection in uv coordinates. The unit of all four parameters is pixels. The u and v directions run across the width and height of the image plane.

In this example fu = 701.4165958679, fv= 701.480279171, cu = 668.2392112416, and cv = 517.9783218077.

- camera: 
    intrinsics: 
      data: [701.4165958679, 701.480279171, 668.2392112416, 517.9783218077]

Distortion model

The equidistant distortion model uses five parameters k1, k2, k3, k4, k5 (see equation 6 of the reference mentioned above). We fix k1 to 1, the other four parameters are encoded in camera/distortion/parameters in the same order.

In this example k1 = 1, k2= -0.0416026702, k3 = 0.0022689289, k4 = -0.0027567794, and k5 = 0.000401603.

- camera:
    distortion:
      parameters: 
        data: [-0.0416026702, 0.0022689289, -0.0027567794, 0.000401603]

Extrinsics

Extrinsic calibration is encoded in the T_B_C/data node as a homogeneous transformation matrix with the translational part in meters. It expresses the pose of the camera in the frame of the IMU.

- T_B_C: 
  data: [
    -0.0089889092,  0.0097254569,  0.9999123037,  0.0520086092,
    -0.9999261367, -0.0082678741, -0.0089086176,  0.0488993112,
     0.0081805087, -0.9999185256,  0.0097990577, -0.011757515,
     0.0,           0.0,           0.0,           1.0]

kalibr_calib.yaml

This file is meant for the stereo demo we provide on github.

Obtaining the calibration files

The above mentioned files can be downloaded with the driver using the following command.

alphasense download_calibration - 7s_sensors.yaml

The command will output the file path on STDOUT. This can be used in a script to automatically get the correct calibration even when switching the Alphasense Core.

CALIBRATION_FILE_PATH=`alphasense download_calibration - 7s_sensors.yaml`
rosrun my_vio_package my_vio_node _calibration_file:=$CALIBRATION_FILE_PATH

The calibration can also be downloaded to a specific path on the computer

alphasense download_calibration - 7s_sensors.yaml ~/my-favorite-path/my-calibration-file