comparative-calibration

Automated data acquisition scripts for Precision Camera Calibration Using Known Target Motions Along Three Perpendicular Axes. You will need this package, the Genuis command-line calculator, cal-3d and its dependencies, and the testing forks of ros-industrial and image_pipeline:

cd ~/ros_ws/src
sudo apt install genius
git clone [email protected]:cwru-robotics/industrial_calibration.git -b indigo-devel
git clone [email protected]:ros-industrial/intelligent_actuator.git
git clone [email protected]:cwru-robotics/image_pipeline.git
git clone [email protected]:cwru-robotics/3d-calibration.git
cd ..
catkin_make

Once everything is set up, you will need to acquire the initial image data that the error generation scripts will permute. Then, you can create the comparative data.

Data Generation

Acquiring Initial CWRU Data

# From comparative-calibration root directory...
roslaunch intrinsic_simulation cam_launch.launch acquire_file:=$PWD/raw_data_cwru/task_description.yml

roslaunch intrinsic_acquisition intrinsic_acquisition.launch path:=$PWD/raw_data_cwru/task_description.yml

roslaunch dotboard_detection dots_detect.launch adf_path:=$PWD/raw_data_cwru/task_description.yml csv_path:=$PWD/raw_data_cwru/intrinsic_detections.csv

Acquiring initial industrial data

# From comparative-calibration root directory...
roslaunch intrinsic_simulation cam_launch.launch acquire_file:=$PWD/raw_data_ind/task_description.yml cam_file:=$PWD/raw_data_ind/camera_headon.yml

roslaunch intrinsic_acquisition intrinsic_acquisition.launch path:=$PWD/raw_data_ind/task_description.yml

Acquiring initial Zhang data

Very similar to CWRU acquisition, but you need to have 3d-calibration package be on branch zhang_gen to get the rotation. Then,

# From comparative-calibration root directory...
roslaunch intrinsic_simulation cam_launch.launch acquire_file:=$PWD/raw_data_zhang/task_description.yml

roslaunch intrinsic_acquisition intrinsic_acquisition.launch path:=$PWD/raw_data_zhang/task_description.yml

# This version of the target detector eliminates illegible images so that the decimation step removes ten ACTUALLY USEFUL elements each run.
roslaunch dotboard_detection dots_detect.launch adf_path:=$PWD/raw_data_zhang/task_description.yml csv_path:=$PWD/raw_data_zhang/intrinsic_detections.csv

Generating comparative data

The various folders in this repository contain scripts called generate.sh and calibrate.sh. If it exists, generate.sh should be run first. calibrate.sh will then output the results of many calibrations to the command line. I use Terminator's logging feature (right click on the terminal and select 'start logging' before running anything) to save all of the output in a single file. I suppose you could also use the > command, but I ran into problems with this not properly interleaving (or skipping entirely) output from ros_info, ros_error, printf, and echo sources.

Also, be aware that some of these acquisition scripts can take a long time to run, on the order of hours or even a day or two.

Processing data

I used grep to strip out all lines that don't contain important data. i.e.:

grep "Read in\|RMS value\|fx =\|k1 =" cwru_scale_raw.txt > cwru_scale_proc.txt

grep "s =\|final cost\|camera_matrix data\|distortion data" ind_target_scale_raw.txt > ind_target_scale_proc.txt

grep "s =\|n =\|D =\|P =\|RPE is" zhang_target_scale_raw.txt > zhang_target_scale_proc.csv

Some additional manual find-and-replace was used to turn the resulting lines into neat columnar data. Additionally, it would be a good idea to give each data set a manual look-over to see if any weird messages happen to be included in it that will make the data difficult to parse.