03_sensor_calibration.md

3. Sensor Calibration

Estimate your intrinsic/extrinsic parameters on your sensor system using tier4/calibration_tools.

Note

Perform the following tasks on the x86-based ECU.

3-1. Calculate intrinsic parameters for cameras

Before starting intrinsic parameters estimation, start camera streaming:

# optional: access Jetson-based ECU remotely
ssh <IP_address(default: 192.168.2.2)> -l <username>

# start camera streaming
cd edge-auto-jetson
source install/setup.bash

ros2 launch edge_auto_jetson_launch edge_auto_jetson.launch.xml perception:=false

Then, launch calibration_intrinsic to estimate your intrinsic parameters of your cameras. See this document for detailed operation on the tool.

cd edge-auto
source install/setup.bash

ros2 launch edge_auto_launch calibration_intrinsic_sample.launch.xml

On the first popup window, select ROS topic to perform calibration on live streaming camera data (In the case of using recorded data in ROSBAG, select ROS bag). Afterwards, select the appropriate board options and board parameters, then click Start:

On the second popup window, select the topic name of the camera you would like to perform intrinsics calibration. This tool can accept compressed topic as the target topic, and the compressed data would be preferred to save the network load of the system. Note that each topic on ROS 2 has its quality of service (QoS). If the actual QoS assigned by a publisher and QoS(s) expected by subscribers are not acceptable combinations, the communication between them will not be issued properly. By default, edge-auto-jetson publishes sensor data with BEST_EFFORT QoS reliability, so here choose it on the drop down list.

After data collection, click calibrate and wait until "Calibration status" turns into idle. If intrinsic parameters are estimated correctly, the reports of the estimation are displayed on some additional popup windows:

Repeat this step for all cameras in your system.

Intrinsic parameters for all cameras consisting of your system (e.g., camera0 and camera1 in this tutorial) should be stored on individual_params. After acquiring the files, put them in appropriate folder on your Jetson-based ECU so that they are loaded and published as camera_info topics:

edge-auto-jetson/src/individual_params/individual_params/config/
└── default # <- this will be an identifier, which refered to as the value of `VEHICLE_ID` environment variable, of your system
    ├── camera0
    │   ├── camera_info.yaml  # <- replace this file with your calculated results
    │   ├── trigger.param.yaml
    │   └── v4l2_camera.param.yaml
    ├── camera1
    │   ├── camera_info.yaml  # <- replace this file with your calculated results
    │   ├── trigger.param.yaml
    │   └── v4l2_camera.param.yaml

3-2. (Optional) Get a correction file from LiDAR

Some product series of LiDARs, including HESAI AT128, have the capability to access their own correction file stored inside them. To get better results, you are encouraged to download the correction file from individual LiDAR, and store it in appropriate folder on the x86-based ECU:

edge-auto/src/individual_params/individual_params/config/
└── default
    └── lidar
        └── at128_default.dat # <- replace this file with your downloaded dat file

3-3. Calculate extrinsic parameters between LiDAR and cameras

Finally, launch calibration_extrinsic that matches your LiDAR setup to estimate your extrinsic parameters between your lidar and cameras. See this document for detailed operation on the tool.

cd edge-auto
source install/setup.bash

ros2 launch edge_auto_launch calibration_extrinsic_xt32_sample.launch.xml
## or
ros2 launch edge_auto_launch calibration_extrinsic_at128_sample.launch.xml

After executing the above launch file, two windows will be appear like below:

You do not need to click any points on the RViz window (i.e., the left-hand window in the figure above). If calibration targets are recognized correctly, markers will appear automatically.

To perform sensor fusion, pose relationships (i.e., extrinsic parameters) between all fused sensors need to be registered in TF, which is the ROS-fashion to represent the relationships.

NOTE: In these sample, there is the assumption that all sensors are fixed and their relative positions do not change.

After calculating extrinsic parameters, put the result in the appropriate files on the x86-based ECU:

edge-auto/src/individual_params/individual_params/config/
└── default
    ├── xt32_to_camera0.json # <- replace this file with your calculated results
    └── xt32_to_camera1.json # <- replace this file with your calculated results