Multi-aperture optical flow estimation for artificial compound eye using non-subsampled contourlet transform and guided filter

Abstract

In this article, a multi-scale optical flow estimation for image taken by artificial compound eye (ACE) is investigated. The optical flow estimation of ACE needs to be adapted by designing algorithms according to its unique multi-aperture characteristics A more general filter for the regularization term, rather than a single iterative solution in the traditional variational model, is devised using the non-subsampled contourlet transform (NSCT) to enforce band-decomposition to estimate optical flow field. To circumvent the spillover and error of the single-aperture fringe flow field, a flow gradient weight is introduced to suppress it and to enhance motion details. Additionally, low-pass subbands adopt Bayes threshold operation with the advantage of efficiently eliminating outliers, and more high-pass subbands adopt guided filter (GF) with the advantage of separating important details from outliers. The prominent feature of the proposed method is that the accuracy of optical flow estimation is improved effectively by eliminating outliers. Eventually, the superiority of explored optical flow estimation is demonstrated through experimental results.