The main objective of this work is to track the aortic valve plane in intra-operative fluoroscopic images in order to optimize and secure Transcatheter Aortic Valve Implantation (TAVI) procedure. This paper is focused on the issue of aortic valve calcifications tracking in fluoroscopic images. We propose a new method based on the Tracking-Learning-Detection approach, applied to the aortic valve calcifications in order to determine the position of the aortic valve plane in intra-operative TAVI images. This main contribution concerns the improvement of object detection by updating the recursive tracker in which all features are tracked jointly. The approach has been evaluated on four patient databases, providing an absolute mean displacement error less than 10 pixels (≈2mm). Its suitability for the TAVI procedure has been analyzed.