Description and Usage Instructions

The Robust Automatic Detection in Laser Of Calibration Chessboards (RADLOCC) toolbox allows for the automatic extrinsic laser-camera calibration. The toolbox provides the RADLOCC algorithm which automatically extracts the calibration board lines from the laser scan. This algorithm was added onto the original Laser-Camera Calibration Toolbox by Fabio Ramos and James Underwood. The calibration process used in the toolbox is based on [1]. The method requires a calibration plane (such as a chessboard) to act as a common dataset between the laser range finder and the camera. The points in the laser scans corresponding to the calibration plane need to be selected. Using the RADLOCC algorithm [2], these points are extracted automatically. The toolbox assumes that the intrinisc parameters already exist. For automatic camera calibration check the RADOCC toolbox.

The toolbox provides the following functionality:

1. Automatic selection of board lines.
2. Automatic guess of initial estimate (manual input of initial estimate is also possible).
3. Optional manual selection of board lines.
4. Extrinsic laser-camera calibration.
5. Calibration error analysis and improvement.

The toolbox provides a GUI for the user's convenience.

The toolbox has the following requirements:

1. The toolbox assumes that camera calibration has already been performed. Refer to the RADOCC toolbox for automatic camera calibration.
2. The toolbox relies on a calibration plane to act as a common dataset between the camera and the laser range finder. It is assumed that the camera calibration provides the orientation and position of the calibration plane.
3. It also requires that changes to the background to remain as minimal as possible. This is required for the automatic selection process.
4. It is recommneded that the calibration board not be rotated more than 45 degrees about its orthogonal axis.
5. Finally, the board's position should be moved around as much as possible throughout the dataset.

After unzipping the file to folder installfolder, add the directory to the Matlab path:

>>addpath('installfolder'); savepath;

To begin the calibration, set the Matlab path to the path containing the laser data and the file 'Calib_Results.mat' output by the camera calibration. Then, type RADLOCC at the Matlab command line.

The following dialog should pop up.

The following steps introduce a quick start to performing a calibration:

1. Press Read data. Enter the file names (Refer to the DATA FORMAT section for details about the format of the calibration dataset). Reading the data may take some time depending on the size of the data. Fortunately, after the data is read, the user can press Save to save the data and use Load next time for quicker loading times.
2. Press Auto select. The program will ask for two threshold values: rough line threshold and fine line threshold. The line extraction stage of the automatic selection process requires two thresholds. The rough threshold is used for obtaining the intial estimate for the calibration and the fine threshold is used for the final optimisation. The selection of these values will depend on the quality of the laser range finder, the default values should be fine for a SICK laser. The program will then ask whether the user wants the program to attempt to guess the initial estimate. Select yes or manually enter the intial estimate. The program will then run an optimisation to find the boards in the laser scans. Finally, the user might be asked to verify some lines suspiciously extracted and/or to manually select the board from laser scans with no boards extracted.
3. Press Calibrate to obtain the calibration results with uncertainty. The uncertainty values represent the estimated standard deviation of the errors.
4. The calibration process should be complete; however, there might be need for refining the calibration. For visual inspection, press Laser onto image to reproject the laser scans onto the images.
5. The Analyse error feature displays the error for each extracted board. This feature can be used with Add/Suppress scans option to remove outlier scans.
6. Verify the results by reprojecting the laser scans onto the images. This can be done by pressing Laser onto image and then selecting the image numbers. What is important to notice is the consistency between the range jumps of the laser scan and edges in the image as shown below.

The calibration results can be adjusted/improved by one of the following methods:

1. Manually selecting some boards in the laser scans at the end of the automatic selection process. The auto selection process might remove some valid board scans. This reduce the amount of data available for calibration. Manually selecting these scans might improve the calibration resutls.
2. Suppressing outlier scans. For various reasons, some scans might appear as an outlier while analysing the errors. Suppressing these scans might improve the results.
3. Changing the line thresholds. The default line thresholds were tuned for SICK lasers. The accuracy of other laser range finders might differ. Therefore, the toolbox might fail to detect laser boards in such scans. Two thresholds are used, rough and fine. The rough threshold is used to permit the program to successfully obtain an initial estimate for the calibration since a fine threshold might remove alot of board points. The fine threshold on the other hand, is used to remove noise from board scans and reduce error in the calibration.
4. By choosing to Manually select (this does not use any automatic extraction) the laser board scans.

The Laser data file is assumed to be in ascii format. Two files are expected. The first file contains the laser range data. The second file contains the timestamps of the images used in the camera calibration. In the first file each row should correspond to a laser scan. The format of each line should be as follows:

<timestamp> StartAngleRads AngleIncrementRads EndAngleRads RangeUnitType NoAngles [Ranges]

The second file contains the timestamps for the images used in the camera calibration sorted in a column.

RangeUnitType

= 1 for mm

= 2 for cm

= 3 for m

= 4 for km

A point in 3D space with coordinate vector P_c with respect to the camera has coordinate vector P_l in the laser frame. The two vectors can be related by the following equation:

P_l=Φ(P_c-Δ)

where Δ is the translation offset and Φ is the rotation matrix defined by the set of three Euler angles φ_x, φ_y and φ_z.

Φ defines the rotation that takes the camera frame to the laser frame. It is a combination of three rotations (the order is important). The first is a rotation about the x axis by the angle φ_x, the second is about the (possibly new) y axis by the angle φ_y and the last is that about the (again possibly new) z axis by the angle φ_z. Δ is simply the coordinates of the vector extending from the camera origin to the laser origin in the camera’s coordinate frame.

If you have a question or a suggestion, or if you find a bug, please email Abdallah Kassir.