Understanding Calibration

edit

The calibration of a camera lens is done, so that the image sensor data distortion can be rectified before being stored as a file

Nop. Undistorting an image may have several applications. Distortions occur so that cameras can achieve bigger fields of view. But when you are performing image processing, normally you want no distortions in the image. Example: Running a face detection algorithm will work better in an undistorted image.

I see everywhere, that a chess board is taken for calibration. But I could never understand how a chess board can help calibrate the camera?

To put it simply, chessboards are used to infer distortion coefficients and other camera intrinsic parameters. Since the chessboard square sizes are known, and their relative positions are also known, there are algorithms that can calculate the transformations that the lens made to bring the chessboard from its normal appearance to the distorted one. Undistorting is just reverting these transformations.

Or is it that the calibration is done so that the rectification can be done later by the user himself?

Yes. Since normally one doesn't get these parameters from camera sellers, you need to calibrate them if you want to use the undistorted images.

Is there any more parameters that needs to be looked, when we have multiple cameras in the setup?

Depends on the application. For example, if you find an object in Camera A and then want to look at that place in Camera B, you need to know the transformation between A and B (the translation and rotation matrix).

If you want to stitch two images in the overlaping area to create a panorama image, you need to find the common points on both images to calculate the homography and use it to transform one of the image to the plane of the other.