Abstract
The past decades have witnessed the notable development of minimally invasive surgery (MIS). The benefits of this modality of surgery for patients are numerous, shortening convalescence, reducing trauma and surgery costs. In this context, robotic assistance aims to make the surgical act more intuitive and safer. In the domain of cardiac MIS, heartbeat and respiration represent two important sources of disturbances. Even though miniaturized versions of heart stabilizers have been conceived for the MIS scenario, residual motion is still considerable and has to be manually canceled by the surgeon. In this thesis, the focus is put on computer vision techniques for estimating the 3D motion of the heart relying solely on natural structures on the heart surface for active compensation of physiological motions. Two main contributions on the subject of motion compensation for robotized cardiac MIS are proposed. The first is a visual tracking method for estimating the 3D deformation of a region of interest on the heart surface. A thin-plate spline model is used for representing the heart surface deformations and a novel efficient parameterization for 3D tracking using stereo endoscopic images is proposed. The method is robust to illumination variations and large tissue deformations. The second contribution is a robust visual tracking method using motion prediction. A time-varying dual Fourier series for modeling the quasi-periodic beating heart motion is proposed. For estimating the parameters of the Fourier series, a probabilistic framework is based on the Extended Kalman filter (EKF) is used. The visual tracking method is integrated in the heart motion prediction framework, creating an unified framework for estimating the temporal motion and spatial deformation of the heart surface. Experimental results have shown the effectiveness of the proposed methods.