An algorithm for extracting corner points of checkerboard Calibration board (from the Camera calibration section of the Open source algorithm Ptamm)

In the field of machine vision, camera calibration is a very basic task. At present, our frequently used calibration tools are OPENCV and MATLAB.

OpenCV's Checkerboard Corner extraction API requires us to specify the horizontal points of the checkerboard in order to extract the corner points. Its main process is to take the picture two value, morphological processing, to find a number of four-sided, to each quadrilateral number, according to the set of quadrilateral and set of vertical points to the missing quadrilateral complement, redundant culling. If a failure is found, change the parameters such as binary and morphological to continue the process until a failed exit condition is found or satisfied. This is the approximate procedure for finding a checkerboard corner point on a OpenCV. The use of OPENCV also requires the calibration of the image collected by the calibration board is complete, and there is a circle of white edge. Want to know more about OpenCV source code can be consulted.

When using the auto-find checkerboard corner function on Matlab, you can extract the checkerboard corner points by not setting the checkerboard horizontal points, and return the horizontal points of the checkerboard. The main process has not been understood.

Ptamm is a AR/VR open source software, which is the application of camera attitude estimation algorithm in virtual reality or augmented reality. It contains a camera calibration tool. This camera calibration tool does not use OPENCV, but it writes its own algorithm for finding the corner of the checkerboard. The algorithm does not require the complete checkerboard on the picture, and the incomplete checkerboard picture can also be used for camera calibration.

The main process is to study the source code for the corner point of the Checkerboard:

Select a pixel on a circle around each pixel of the picture to determine if the pixels on the ring meet 4 times, whether the swap interval is almost equal, and whether the center pixel value is about the same as the average of all the pixels on the ring; the Jorai point gets to a series of alternate points of the checkerboard corner. Next, select a point in this heap that is closest to the center of the image as the origin of the checkerboard, determine the x, y direction, and optimize the center point of the point, and then, based on the X, y direction, find more checkerboard points in the alternate points, and then optimize the further expansion until the end.

The center and X, y directions are optimized using a Checkerboard corner template (generated from an iterative initial value, modified during iteration) to match the pixels of the same size neighborhood as the picture at that point, matching errors are used to modify the center and X, y directions. If the point's match is still large after several iterations, the point is discarded.

These are just the general process. What I saw a long time ago is written by memory.