You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardexpand all lines: atom_examples/rgb_rgb_system/README.md
+59-31
Original file line number
Diff line number
Diff line change
@@ -42,52 +42,80 @@ This will put the bag file into your $ROS_BAGS folder.
42
42
43
43
# Calibration
44
44
45
-
...
45
+
The calibration of any robotic system using **ATOM** may have several variants. We recommend a careful reading of the [documentation](https://lardemua.github.io/atom_documentation/) to learn all the details.
46
46
47
-
<!--
48
-
## Configuring a calibration package
49
-
Once your calibration package is created you will have to configure the calibration procedure by editing the softbot_calibration/calibration/config.yml file with your system information. Here is an example of a config.yml file.
In this section, out goal is to carry out the simplest possible calibration pipeline for the **rgb_rgb_system**.
52
48
53
-
After filling the config.yml file, you can run the package configuration:
49
+
To calibrate, we will need a bagfile called [rgb_rgb_system_example_bag.bag](https://drive.google.com/file/d/1Noo3eZh72m-xRobYZywdo1wtqg7e4wGa/view?usp=sharing), which contains a recording of the system's data when viewing a calibration pattern in several positions.
50
+
We produced the bagfile by bringing up the system and then recording a bagfile as described above.
51
+
This is a small bagfile with 40 seconds / 60MB for demonstration purposes. Typically, calibration bagfiles are larger.
Download the bagfile and put it in **$ROS_BAGS/rgb_rgb_system**.
56
54
57
-
This will create a set of files for launching the system, configuring rviz, etc.
55
+
Next we describe each of the steps in the calibration pipeline.
58
56
59
-
## Collect data
57
+
## Creating a calibration package
60
58
61
-
To run a system calibration, one requires sensor data collected at different time instants. We refer to these as data collections. To collect data, the user should launch:
59
+
See also the [generic documentation](https://lardemua.github.io/atom_documentation/procedures/#create-a-calibration-package) on this topic.
**NOTE**: This procedure is carried out only once, and was done already. As such, this ros package is already [included in the atom repo](https://github.com/lardemua/atom/tree/miguelriemoliveira/issue629/atom_examples/rgb_rgb_system/rgb_rgb_system_calibration). Therefore, you **should not execute this instruction** for the rgb_rgb_system.
This is the [config.yml](https://github.com/lardemua/atom/blob/miguelriemoliveira/issue629/atom_examples/rgb_rgb_system/rgb_rgb_system_calibration/calibration/config.yml) that we wrote to define the calibration. There are two sensors to be calibrated, named **rgb_left** and **rgb_right**. The pattern is a charuco marker.
72
+
The configuration file points to the bagfile mentioned above, and the _anchored_sensor_ is defined as the **rgb_left** sensor.
79
73
80
-
To evaluate the calibration that was done, its need to do the annotation
We will use as example the [rgb_rgb_system_example_dataset](https://drive.google.com/file/d/1i4cOp1BcPsifmr2VLpYM0mYa0_SWaUGK/view?usp=sharing), which contains 4 collections, as shown bellow.
90
+
91
+
Download and decompress the dataset to **$ATOM_DATASETS/rgb_rgb_system/rgb_rgb_system_example_dataset**.
This will produce a table of residuals per iteration, like this:
112
+
113
+
114
+
115
+
This is the table presented once calibration is complete, which shows average reprojection errors of under 1 pixel. Sub-pixel accuracy is considered a good result for rgb camera calibration.
116
+
117
+
Since this is a simulation, the original pose of the cameras is actually the one used by gazebo to produce the images. This means that the cameras are already positioned in the actual ground truth pose, which means that the calibration did not do much in this case. In a real system, the sensors will not be positioned at the ground truth pose. In fact, for real systems, we do not know where the ground truth is.
118
+
119
+
To make sure this ATOM is actually calibrating sensor poses in simulated experiments, we use the --noise_initial_guess (-nig) flag. This makes the calibrate script add a random variation to the initial pose of the cameras, to be sure they are not located at the ground truth:
0 commit comments