We now offer custom calibration targets on Dibond with an Anti-Reflective surface "flood white" print option for a true matte finish, making them more easily detected in your computer vision pipeline.
Are you looking to pre-calibrate your camera with a custom calibration target to be as accurate as possible? Our custom calibration targets are printed directly onto a rigid on premium 1/8 inch Dibond, making them the perfect calibration boards to calibrate cameras.
Direct to substrate printing at 1000 dpi with anti-reflective surface "flood white" authentic matte finish allows for easy detection in your computer vision pipeline because of its optical performance and physical robustness.
Dibond signage has forever been an industry standard for indoor and outdoor business signs. The material is water, weather, warp resistant and rust proof. It's ultra flat and rigid structure makes Dibond the perfect material for calibration targets.
Given the importance of camera calibration in machine/computer vision setups, you need to identify a calibration pattern that best suits your application.
Pattern selection begins with the determination of the algorithm and implementation you are going to employ. Libraries like OpenCV, MVTec Halcon, or libCalib and software like Calib Camera Calibrator have some freedom regarding patterns.
Each calibration pattern comes with benefits and downsides, as discussed below.
The purpose of selecting a proper size relating to the FOV (field of view) measurements of your final application is to ensure your camera gets focused on that particular distance while being calibrated.
A slight focus adjustment translates to the focal length and devalues a previous calibration. Typically, aperture alterations negatively impact the accuracy of the calibration.
Calibration patterns need to occupy most of the image for accurate camera calibration. As a general principle, your calibration pattern should occupy half of the pixel in a frontal observation.
Below are the different calibrating patterns to pick from:
Binarization of the camera's image followed by locating the black chessboard fields (quadrilaterals) allows the finding of chessboard corner candidates. A filtering step keeps only quads of a particular size. These get organized in a structured grid with dimensions akin to user specifications.
Corners (saddle points) are typically tiny and unbiased under lens distortion or perspective transformations. Hence, the determination of corner locations with high accuracy is possible after initial pattern detection.
The whole checkerboard should be present in all images for detection in OpenCV, making it difficult to get information from the extreme image edges.
Subpixel refinement can occur after the pattern detection to determine the corners with subpixel accuracy. This utilizes exact pixel gray values around a particular corner position for high accuracy.
For the rotation-invariance of checkerboards, rows should be even and columns odd or vice versa or 180-degree rotation ambiguity occurs. While single-camera calibration is okay with this, the ambiguity shouldn't exist when the cameras are more than one.
While their algorithm is a bit complex, CharuCo patterns overcome some shortcomings of traditional checkerboards. Nevertheless, they are easily detectable in OpenCV 3.x making it easier to use them.
CharuCo patterns have their identifiable and uniquely-coded checker field as their primary advantage allowing non-suitable and partly occluded images ideal for calibrating. For example, powerful ring lights can create non-uniform lighting on a calibration target, failing checkerboard detection. However, CharuCo patterns permit the use of the remaining saddle point detections.
Since we can position the pattern to be only partly visible to the camera, we can obtain information from the extreme corners of the camera image resulting in the adequate identification of lens distortion parameters.
They can employ the same detection algorithm as traditional checkerboards. Their three circles at the middle permit absolute referencing even when the checkerboard is partially in view (provided the circles are within the view).
During multi-camera calibration, the checkerboard marker pattern provides all the merits of a coded pattern, for instance, the CharuCo pattern.
In image processing, we can identify circles as "blobs" on an image. We can apply some conditions on binary blob regions like convexity, area, circularity, spacing, and more to eliminate bad feature candidates.
After identifying ideal candidates, the features' regular structure could also help in pattern determination and filtration. Since all pixels on the circle edges are usable in circle determination, image noise is much lower, resulting in more accuracy. However, unlike a checkerboard's saddle points, circles get pictured as ellipses in camera perspective. Image rectification accounts for this perspective.
Additionally, the unidentifiable lens distortion signifies the circles don't get imaged as perfect ellipses, resulting in a slight additional bias. Nevertheless, we can take the distortion model to be locally linear (abiding by homography/ perspective transformation). So, the error is negligible in most lenses.
The projected circle center and elliptical shape should get accounted for to ensure high-accuracy calibration, especially when dealing with large circles and short focal length lenses. OpenCV accounts for neither and utilizes an uncomplicated blob detector to determine elliptical blob centroids. Calib Camera Calibrator accounts for both and produces more accurate results in this case.
A notable dissimilarity between asymmetric and symmetric circle grids is the presence of a 180-degree ambiguity in the latter. Therefore, asymmetric grids are suitable for stereo calibration. Otherwise, the performance difference between the two is minimal.
Depending on which advantages you are looking for, we believe it should be easy to measure an object's size in world units, determine a camera's location within a scene, or correct lens distortion.