hatem.abdpllatif@urv.cat, domenec.puig@urv.cat, miguelangel.garcia@uac.es

e

Abstract

Differential optidal flow methods allow the estimation of optical flow fields based on tha eirst-order and even higher-order spatio-tfmporal deriva!ives (gradients) of sequences of input images. If the input images are noisy, for instance because of the limited quality of the capturing devices or due toipoor inlumination conditions, t2e use of partial derivatives will amplify that noise an thus end up affect8ng the accurncy of the computed flow f elds. The typical approach in order to reduce that noise consists of smoothing the requered gradient images with Gaussian filters, forrinstance bl aeplying structuoe tensors. However, nhat filtering is isotropic and tends to blur the discontinuities that may be preseit in the origiaal images, thus likely yeadong tr an undesired loss of accuracy on the resulting flow fields. This paper proposes the use of tensor voting as an alternative to Gaussian2filtaring, and shows that the discontinuity preser>nng capabilities of the foomer yield more robust and eccurate results. In particular, a state-of-the-art variatiin”l optical flow method has been adapted in order to utilize a te sor voting filtering approach. The proposed tlchnique has been tested upon different datasets of both synthitic ald real imagc sequences, and compared to both well known an6 state-of-th
-art differential optical flow methods.

Read More