Figure - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Quantitative comparison of two-frames and three- frames using the endpoint error metric with different masks.
Source publication
Approaches that use more than two consecutive video frames in the optical flow estimation have a long research history. However, almost all such methods utilize extra information for a pre-processing flow prediction or for a post-processing flow correction and filtering. In contrast, this paper differs from previously developed techniques. We propo...
Contexts in source publication
Context 1
... 6 illustrates this experiment and shows that the filtering operation in the cost volume can improve the discrete optical flow estimation. Table 3 gives the quantitative evaluation results. One can see that for the MPI-Sintel dataset the accuracy of our estimation without the backward flow consistency check increases up to 50% in occluded regions and up to 9% in non-occluded regions in comparision with the two-frame approach, which are the original results of DCFlow algorithm before consistency check. ...Context 2
... can see that for the MPI-Sintel dataset the accuracy of our estimation without the backward flow consistency check increases up to 50% in occluded regions and up to 9% in non-occluded regions in comparision with the two-frame approach, which are the original results of DCFlow algorithm before consistency check. In the case of the cost volume pre-filtering accuracy of the final result (three-frame+ in Table 3) increases further by 21% and 12% in non-occluded and occluded regions respectively. In this experiment we also leverage a variational energy minimization post-processing method [49] to obtain our final optical flow results (three-frame++ in Table 3). ...Context 3
... the case of the cost volume pre-filtering accuracy of the final result (three-frame+ in Table 3) increases further by 21% and 12% in non-occluded and occluded regions respectively. In this experiment we also leverage a variational energy minimization post-processing method [49] to obtain our final optical flow results (three-frame++ in Table 3). In comparison with the DCFlow results after interpolation and post processing (Table 5), our method reaches the same accuracy level in non-occluded regions directly without the consistency check and interpolation. ...Citations
... Furthermore, Chen et al. [58] presented a segmentation-based patch matching framework to cope with the issue of over-segmentation. Yang et al. [59] proposed a maximum likelihood function calculation method which increases the robustness of the matchingbased optical flow estimation methods. ...
Recently, the interpolation of correspondences method has been widely used in optical flow estimation, because it produces an accurate flow field and costs little runtimes. However, most of the existing matching-based optical flow methods are usually susceptible to non-rigid motion and large displacements. We propose in this article a large displacement optical flow estimation method based on robust interpolation of sparse correspondences, named Riscflow. First, we utilize the deep matching model to achieve an initial matching result of two consecutive frames, and then we exploit a grid-based motion statistics optimization scheme to remove the outliers from the initial matching field. Second, we propose a random forest-based motion boundary extraction model and construct a sparse-to-dense interpolation method by using the boundary information to prevent the dense matching field from edge-blurring. Third, we design a global optical flow estimation method by using an energy function to optimize the dense matching field. Finally, we respectively run the proposed method on the MPI-Sintel and UCF101 databases to conduct a comprehensive comparison with some state-of-the-art optical flow approaches including the variational methods, the matching-based methods, and the deep learning-based methods. The comparison results demonstrate that the proposed method has high accuracy and good robustness of optical flow estimation, and especially gains the benefit of edge-preserving under non-rigid motion and large displacements.
Motion estimation is an essential technique to be carried out during the video encoding, characterized by the complex inter-frame prediction challenges. A review of existing approaches exhibits that there is still enormous scope for improvement. Therefore, the proposed system introduces a unique and straightforward process for performing inter-frame prediction by applying an enhanced exhaustive search mechanism followed by a simplified machine learning scheme. The further encoding process is carried out with respect to quantization parameters, frame size, and intra block size. The simulation carried out in MATLAB shows that the proposed system offers a good signal quality and response time compared to H.265 and H.264.
