As we all know, in machine vision, image measurement, photogrammetry, 3D reconstruction and other applications, the use of calibration plates is often inseparable. However, with the continuous development of the times, different calibration boards have been introduced in these years. Each calibration board has unique properties and benefits. Therefore, it is very important to learn to choose the type of calibration board. What are the types of calibration plates? Here's a brief introduction:
Type of calibration plate:
Type 1: Checker
This is a very popular and common pattern design. Usually, the candidate points of chessboard corners are found by first binarizing the camera image and finding the quadrangle (black chessboard area). In the filtering step, only the quadrangles that meet the specific size criteria are retained and organized in a regular grid structure whose size matches the size specified by the user.
After the preliminary detection of the calibration plate, the corner position can be determined with very high accuracy. This is because the angle (mathematically: saddle point) is basically infinitesimal, so it is unbiased under perspective transformation or lens distortion.
In OpenCV, the entire chessboard must be visible in all images to be detected. This usually makes it difficult to obtain information from the edge of the image. These regions are usually good sources of information because they appropriately constrain the lens distortion model.
After the checkerboard is detected, sub-pixel thinning can be performed to find saddle points with sub-pixel accuracy. This makes use of the exact gray value of the pixels around the given corner position, and the accuracy is much more accurate than that allowed by the integer pixel position.
An important detail about the checkerboard target is that in order to keep the rotation unchanged, the number of rows must be even, and the number of columns must be odd, or vice versa. For example, if both are even numbers, there is ambiguity about 180 degrees of rotation. For the calibration of a single camera, this is not a problem, but if the same point needs to be recognized by two or more cameras (for stereo calibration), this ambiguity must not exist. This is why our standard chessboard targets have even/odd rows/columns.
Type 2: Round grid
Circular grid is also a popular and very common calibration target design. It is based on a circle, or a white circle on a white background, or a black (black) circle on a white background. In image processing terminology, a circle can be detected as a "spot" in an image. Some simple conditions, such as area, roundness and convexity, can be applied to these binary speckle regions to remove candidate bad feature points.
After finding suitable candidates, the pattern is recognized and filtered again using the rule structure of features. The circle can be determined very accurately because all the pixels around the circle can be used to reduce the impact of image noise. However, unlike the saddle point on the chessboard, a circle is imaged as an ellipse from the camera's perspective. This view can be explained by image correction. However, unknown lens distortion means that the circle is not a perfect ellipse, which adds a small offset. However, we can regard the distortion model as piecewise linear (subject to perspective transformation/homography), so this error is very small in most lenses.
An important difference between symmetric and asymmetric circular grids is that symmetric circular grids have a 180 degree fuzziness, as explained in the "chessboard" section. Therefore, asymmetric meshes are necessary for stereo correction. Otherwise, the performance of both types will not be much different.
Type 3: CharuCo
The CharuCo calibration board overcomes some limitations of the traditional chessboard. However, their detection algorithms are a bit complex. Fortunately, CharuCo detection is part of the OpenCVs contrib library (starting with OpenCV 3.0.0), which makes it easy to integrate this advanced method.
The main advantage of CharuCo is that all light inspector fields are uniquely encoded and recognizable. This means that even partially occluded or non ideal camera images can be used for calibration. For example, strong annular light may produce uneven illumination (semi specular reflection area) on the calibration target, which will lead to the failure of ordinary checkerboard detection. With CharuCo, the remaining (good) saddle point detection can still be used. Saddle point positioning can be refined using sub-pixel detection like a chessboard.
This is a very useful attribute for viewing areas close to the corners of the image. Since the camera can only see a part of the target due to its location, we can collect information from the edges and corners of the camera image. This usually leads to very good robustness in determining lens distortion parameters. Therefore, we strongly recommend using the CharuCo calibration board. OpenCV 3. x is available.
Naturally, CharuCo targets can be used for stereo calibration. In this case, some code needs to be executed to find the points detected separately in each camera, as well as the points detected in both cameras (intersections).
To sum up, there are three common types of calibration plates. In the process of selecting calibration plates, we must choose the correct type. Because calibration plates have certain degrees of freedom, they have their own advantages and limitations, so we must choose the appropriate type of calibration plates according to our actual needs.
