Modification of the optical flow horn-schunck estimation incorporating exaustive search

Optical Flow is the pixels apparent movement pattern of two consecutives images. There are differents methods to estimate the optical flow: variational methods and exaustive methods. Exhaustive methods take a neighborhood around a point in the first image of the sequence and search for the most similar in the second image. This procedure is known as correspondency estimation. This work evaluate the effect of the incorporation of new correspondecies to a classical variational method to estimate the optical flow. With this goal a variational model is stated. This variational model considers a function k(x) to decide if a correspondency is incorporated or not to the variational model. Obtained results in a standar database show that the incorporation of new correspondencies improve the performance of the optical flow classical method. The obtained result combines the precision of the variational method with exhaustive search. This method was implemented in a GPU satisfactorily, and reducing the processing time to a few seconds.