In several medical applications such as liver or kidney biopsies, an anatomical region needs to be continuously tracked during the intervention. When using ultrasound (US) image modality, tracking soft tissues remains challenging due to the deformations caused by physiological motions or medical instruments, combined with the generally weak quality of the images. In order to overcome the previous limitation, different techniques based on physical model have been proposed in the literature. [SMSM06] proposed a registration method based on the mass-spring system in order to constrain the deformation, and Zhang et al [ZW13] introduced an other registration technique based on finite element model where the extraction of the scale invariant features is needed. However, their model are built from features which are difficult to extract in US images due to the speckle noise. Finally, Marami et al [MSFC14] presented very recently an elastic registration method applicable to multi-modality image registration where the deformation is computed from modality independent neighborhood descriptor. In this paper, we propose an approach for tracking deformable target within 2D US images based on a physical model driven by smooth displacement field obtained from dense information. This allows to take into account highly localized deformation in the US images. Section 2 presents our method based on a combination of an intensity-based approach and a physically-based model. Section 3 describes the performances of our approach and comparisons on real data. Section 4 concludes the paper.