Camera Calibration: Dimensioning your Target and Acquiring good Images
In addition to having a high-quality calibration target, capturing images for camera calibration is a very critical step to accurate calibration performance that is also highly error-prone.
In this post we will highlight how to avoid common beginner mistakes, present our best practices for selecting the correct calibration target and explain how to achieve optimal images for camera calibration.
The Calibration Target
The purpose of a calibration target is to define a 3-D world coordinate system of known points on the calibration object. These known points can be detected by algorithms from common calibration libraries/software (e.g. OpenCV, MATLAB) that allow to relate the observed points [pixel] from the camera to 3-D world coordinates [mm] on the calibration object and subsequently estimate the camera parameters. The known points on a calibration target, also called features, are defined by various patterns, each with its unique properties. Choosing the right pattern depends, among other things, on the calibration library used and will be covered in another blog post.
Below are some guidelines for selecting the right dimension and feature resolution of your calibration target:
- The physical size of your
desired calibration pattern is an important feature and depended on the
cameras measurement field of view. If you place your calibration board
central in the cameras measurement field of view, it should cover at least 0.25x the size of your image when looking at the target frontally.
For parameter constrain during calibration, it is beneficial if the
calibration target occupies as much of the image as possible, but you
also have to angle the target with respect to the focal axis without
leaving the depth of field in some images (more on this later). With
this in mind, we recommend to chose
your physical target size to be between 0.25-0.75x the size of your
image when looking at the target from a fronto-parallel setup.
- The appropriate number of
features on your calibration target depends on the resolution of your
camera and the illumination settings during image capture. In general, a
high number of features on the target leads to better calibration
results if the resolution of the camera is sufficient. We have therefore
developed two variants for each pattern type with different resolutions
- medium and high. We recommend to use
the high resolution targets if you have a controlled lighting setup
and/or a camera resolution above 2 Megapixels.
- Keep in mind: The calibration is mostly only as accurate as the calibration target used. We recommend using custom printed targets for testing and validation purposes only, as humidity and temperature can affect the dimensions and flatness of your custom made target and therefore worsen your calibration results.
- The physical size of your
desired calibration pattern is an important feature and depended on the
cameras measurement field of view. If you place your calibration board
central in the cameras measurement field of view, it should cover at least 0.25x the size of your image when looking at the target frontally.
For parameter constrain during calibration, it is beneficial if the
calibration target occupies as much of the image as possible, but you
also have to angle the target with respect to the focal axis without
leaving the depth of field in some images (more on this later). With
this in mind, we recommend to chose
your physical target size to be between 0.25-0.75x the size of your
image when looking at the target from a fronto-parallel setup.
Image Acquisition
As mentioned above, common calibration procedures need 3-D world points and their corresponding 2-D image points, in order to estimate the parameters of a camera. These correspondences can be acquired by capturing multiple images of a planar camera calibration target. The quality of those data points in the image correlate directly with the outcome of the camera calibration result.
Best practices on how to capture optimal images for camera calibration are listed below:
- Capture the images with a fixed focus (the focus of your camera is determined by your application). Auto-focus must be disabled, as this changes the lens setup and therefore some of your camera parameters.
- Do not change the zoom setting (the zoom is determined by your application), as this changes the focal length.
- Let the camera be stationary and move the calibration target when taking images for calibration.
- Better one too many than too few - take enough images for calibration, sorting them out is easier than retaking them. At least 10 good images should be used for a proper calibration.
- The entire measurement volume should be covered by the acquisition of the images.
- Capture images where the pattern is near to the edges of the image, as lens distortion is not always uniform across the image and increases radially from the center of the image. This is important, otherwise your parameters will not be valid close to image borders.
- To estimate the focal
length correctly, the camera must record perspective distortion. To do
this, the planar calibration target must be angled significantly
(vertically and horizontally) with respect to the focus axis. An angle of ±45° should not be exceeded, otherwise errors will occur due to inaccuracies in feature detection.
- Avoid vibration and motion blur when capturing images. Ideally, the camera and calibration board are each mounted on a rigid suspension (e.g. tripod) and are then positioned relative to each other.
- Illumination is important!
The calibration target should ideally be illuminated from a diffuse
light source. Strong point light sources lead to uneven illumination
and/or shadows, which causes the detection of the features to lose
significant accuracy and the dynamic range of the camera is not used
very well.
By following these practices for target dimensioning and image acquisition, you'll get the most out of your camera calibration.
Have any additional questions or comments regarding this topic?
Ask us: https://calitar.com/Information/Contact/