2015/08/20 by Christian Bailer, Bailer, Christian, Bertram Taetz +3 · 2 citations
Computer Science · Medicine · #Advanced Image Processing Techniques #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #I.4.8 #Retinal Imaging and Analysis
paper · pdf · doi:10.48550/arxiv.1508.05151
openalex publication_date 2015/08/20 · openalex created_date 2022/10/03 · openalex updated_date 2026/07/28
Modern large displacement optical flow algorithms usually use an\ninitialization by either sparse descriptor matching techniques or dense\napproximate nearest neighbor fields. While the latter have the advantage of\nbeing dense, they have the major disadvantage of being very outlier prone as\nthey are not designed to find the optical flow, but the visually most similar\ncorrespondence. In this paper we present a dense correspondence field approach\nthat is much less outlier prone and thus much better suited for optical flow\nestimation than approximate nearest neighbor fields. Our approach is\nconceptually novel as it does not require explicit regularization, smoothing\n(like median filtering) or a new data term, but solely our novel purely data\nbased search strategy that finds most inliers (even for small objects), while\nit effectively avoids finding outliers. Moreover, we present novel enhancements\nfor outlier filtering. We show that our approach is better suited for large\ndisplacement optical flow estimation than state-of-the-art descriptor matching\ntechniques. We do so by initializing EpicFlow (so far the best method on\nMPI-Sintel) with our Flow Fields instead of their originally used\nstate-of-the-art descriptor matching technique. We significantly outperform the\noriginal EpicFlow on MPI-Sintel, KITTI and Middlebury.\n