Manual calibration for all sensors#

Overview#

In this section, we will make a manual extrinsic calibration of our sensors. We set the sensor poses from measurements or CAD data, and then we correct them by hand while we look at the sensor data in RViz. After this process, we won't get accurate calibration results for final use, but we will have an initial calibration for other tools. For example, in the lidar-lidar or camera-lidar calibration phase, we will need initial calibration for getting accurate and successful calibration results.

We need a sample bag file for the calibration process which includes raw lidar topics and camera topics. The following shows an example of a bag file used for calibration:

ROS 2 Bag example of our calibration process

Files:             rosbag2_2023_09_06-13_43_54_0.db3
Bag size:          18.3 GiB
Storage id:        sqlite3
Duration:          169.12s
Start:             Sep  6 2023 13:43:54.902 (1693997034.902)
End:               Sep  6 2023 13:46:43.914 (1693997203.914)
Messages:          8504
Topic information: Topic: /sensing/lidar/top/pointcloud_raw | Type: sensor_msgs/msg/PointCloud2 | Count: 1691 | Serialization Format: cdr
                   Topic: /sensing/lidar/front/pointcloud_raw | Type: sensor_msgs/msg/PointCloud2 | Count: 1691 | Serialization Format: cdr
                   Topic: /sensing/camera/camera0/image_rect | Type: sensor_msgs/msg/Image | Count: 2561 | Serialization Format: cdr
                   Topic: /sensing/camera/camera0/camera_info | Type: sensor_msgs/msg/CameraInfo | Count: 2561 | Serialization Format: cdr

Extrinsic manual calibration#

Note

Earlier versions of CalibrationTools had an extrinsic_calibration_manager package with a manual calibrator. The current CalibrationTools does not have this package. Its tunable_static_tf_broadcaster package (an rqt_reconfigure tool to change a tf by hand) is not built by default, because the package has a COLCON_IGNORE file. So this page uses only the files of your sensor kit description and RViz.

Setting the initial values#

Open sensor_kit_calibration.yaml and sensors_calibration.yaml in the config/ folder of your sensor kit description package. Write the pose of each sensor from your measurements or from the CAD data of your vehicle. The creating sensor model page explains the frames in these files.

For tutorial_vehicle, there are four sensors in sensor_kit_calibration.yaml (two lidars, one camera, one GNSS/INS):

sensor_kit_base_link:
  rs_helios_top_base_link: # Lidar
    x: 0.0
    y: 0.0
    z: 0.0
    roll: 0.0
    pitch: 0.0
    yaw: 0.0
  rs_bpearl_front_base_link: # Lidar
    ...
  camera0/camera_link: # Camera
    ...
  GNSS_INS/gnss_ins_link: # GNSS/INS
    ...

Checking the calibration in RViz#

Build the sensor kit description package after each change:

cd <YOUR-OWN-AUTOWARE-DIRECTORY>
colcon build --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=Release --packages-select <OWN-SENSOR-KIT>_description

Then publish the tf of your vehicle and sensors with tier4_vehicle_launch. It starts robot_state_publisher with the URDF of your vehicle model and sensor model:

ros2 launch tier4_vehicle_launch vehicle.launch.xml vehicle_model:=<OWN-VEHICLE-MODEL> sensor_model:=<OWN-SENSOR-KIT> launch_vehicle_interface:=false use_sim_time:=true

For tutorial vehicle:

ros2 launch tier4_vehicle_launch vehicle.launch.xml vehicle_model:=tutorial_vehicle sensor_model:=tutorial_vehicle_sensor_kit launch_vehicle_interface:=false use_sim_time:=true

Then play ROS 2 bag file:

ros2 bag play <rosbag_path> --clock -l -r 0.2 \
--remap /tf:=/null/tf /tf_static:=/null/tf_static # if tf is recorded

Start rviz2, set the fixed frame to base_link, and add a PointCloud2 display for each lidar topic. For the camera, add a Camera display. It shows the lidar points on the camera image.

If the point clouds of two lidars do not overlap, change the pose of one lidar.
If the ground is not flat or not at the height zero of base_link, change z, roll, and pitch.
If the lidar points do not fall on the same objects in the camera image, change the pose of the camera.

robot_state_publisher reads the calibration files only when it starts. So after each change, build the package again, and launch vehicle.launch.xml again.

When the sensors are approximately aligned, keep the values in sensor_kit_calibration.yaml and sensors_calibration.yaml. The other calibration tools start from these values.

Warning

The initial calibration process can be important before the using other calibrations. We will look into the lidar-lidar calibration and camera-lidar calibration. At this point, there is hard to calibrate two sensors with exactly same frame, so you should find approximately (it not must be perfect) calibration pairs between sensors.

For camera-lidar pairs, you can also use the interactive camera-lidar calibrator. You select pairs of points in the image and in the point cloud, and the tool calculates the pose of the camera.