The calibration of continuous glucose monitoring (CGM) devices is clinically important due to its role in accurate glucose supervision for diabetic patients. However, the calibration is often performed by finger pricks measurement, which adds patients discomfort while already suffering from diabetes. Therefore, reducing the frequency of finger pricks with undiminished CGM sensor accuracy is essential to improve the patients living condition. In this paper, we aim to propose a fully automated approach for CGM sensor calibration and reduce finger prick frequency. A novel framework is designed and we utilize an unknown input observer to delineate the Blood Glucose (BG)-Interstitial Glucose (IG) relation. We also propose a new cost function with a regularization parameter for sensor parameter estimation. The proposed algorithm is evaluated on the FDA-accepted UVA/Padova T1DM simulator and compared with the original calibration. The meals are designed with various volumes of carbohydrate to simulate the patients' daily intake. The optimal result among all scenarios using mean absolute relative deviation (MARD) metric is 0.10% for 1 calibration (cal) per day while 3.39% for original calibration on the same frequency. After reducing the finger pricks frequency, the results only increase 0.01% and 0.04% using MARD metric for 1 cal/2 days and 1 cal/4 days respectively.