Beating heart motion prediction for robust visual tracking

In the context of minimally invasive cardiac surgery, robotic assistance has significantly helped surgeons to overcome difficulties related to the minimally invasive procedure. Recently, techniques have been proposed for active canceling the beating heart motion for improving the accuracy of the surgical gestures. In this scenario, computer vision techniques can be applied for estimating the heart motion based solely on natural structures on the heart surface. However, visual tracking is complicated by the particular lighting conditions and clutter (smoke, liquids, etc) during surgery. Another challenging problem are the occasional occlusions by surgical instruments. In order to overcome these problems, we exploit the quasi-periodicity of the beating heart motion for increasing the robustness of the visual tracking task. In this paper, a novel time-varying dual Fourier series for modeling the quasi-periodic beating heart motion is proposed. For estimating the series parameters, an Extended Kalman Filter (EKF) is used. The proposed method is applied in a visual tracking task for bridging tracking disturbances and automatically reestablish tracking in cases of occlusions. The efficiency of the prediction method and the sensible improvements in the visual tracking task are demonstrated through in vivo experiments.