Camera calibration is a necessary step in 2D and 3D computer vision. It an important step that needs to be implement in terms of achieving higher accurate results for many computer vision applications such as object detection, tracking, and depth (distance) estimation. Extracting metric information from the hardware camera system is a very extensively and an expensive step. Instead, a non-parametrical approach is needed. In this paper, we proposed a fully automated unsupervised learning approach for multi-camera calibration system. Our approach relies on using different computer vision tools such as image registration and multiscale image fusion to propose two models: localized and globalized calibration models. Our proposed system shows a very promising results when it comes to the calibration results. Different evaluation metrics have been implemented such as SNR, PSNR, MSE, SSIM, and correlation to test our system. The experimental results show that the calibrated videos have gain higher similarity metrics comparing with the uncalibrated ones. In general, the average values of the SSIM and correlation of the calibrated videos have been increased by (28%) and (17%) respectively which gives an indication that our proposed non-parametrical automated calibration system is a generative model which can be used in different hardware multi-camera systems.