Optical flow methods are often used in image processing, for example for object recognition and image segmentation. Traditional optical flow methods use numerical methods, assuming intensity constancy of pixels’ movements. In this work we describe a probabilistic method of modeling the optical flow problem, and discuss the use of Gibbs sampling for optimization of the computed optical flow vector field. In experiments involving test images as well as medical image slices through the short-axis of the left ventricle of the heart, our probabilistic method is compared with the classic Horn-Schunck optical flow method. We demonstrate that our proposed approach probabilistic optical flow method is robust to changes in the shape and intensity of objects tracked. This is a useful property when identifying cardiac structures from time-resolved medical images of the heart, where the shape of the cardiac structures change between consecutive temporal frames of the cardiac cycle.