Recently, model-based visual tracking techniques using endoscopic images have been proposed for estimating beating heart motion and thus actively compensating motion in robot-assisted minimally invasive surgery. However, most existing methods suffer the problems of computation burden with increasing size of target region and number of parameters, and also limited capability to handle partial occlusions. In this paper, an overlapped thin-plate spline approach is proposed. By dividing the target region into overlapped sub-regions, overall computational burden is greatly reduced and robustness against environment disturbances is improved. With the proposed tracking scheme, 3D coordinates of point of interest on beating heart surfaces can be extracted and used to compensate heart motion by a surgical robot, which presents a step further towards successful robotic beating heart surgery using a model-based visual tracking technique. The performance of the proposed tracking method is demonstrated through experiments on phantom data and in vivo videos recorded by the da Vinci surgical platform.