Abstract: This paper introduces a new efficient method for finding the optical flow for a special type of motion and smoothness constrain. The optimisation is mapped into a maximum-flow problem in a graph, which is solved efficiently and leads to an optimal solution. Previously, similar methods were used to solve stereo correspondence [11, 9] and maximum a posteriori estimation problem in Markov random fields [2, 3, 6, 10].
After a general introduction to optical flow, we present a modified version of Horn and Schunck's method [7], then we present how we map the optical flow optimisation problem into a flow graph. Experimental results on simu-lated data are then provided and compared to the modified Horn and Schunck's method. They clearly show the po-tential of our formulation with normalised errors of 9.90with our method vs. 19.87method. Both algorithms were also tested with realistic data in cineangiography of artery to assess blood flow. Again, our method shows a better velocity profile assessment (1.97vs. 8.32the maximum-flow formulation is not iterative and leads to a global optimum.