Coarse to over-fine optical flow estimation

Abstract

We present a readily applicable way to go beyond the accuracy limits of current optical flow estimators. Modern optical flow algorithms employ the coarse to fine approach. We suggest to upgrade this class of algorithms, by adding over-fine interpolated levels to the pyramid. Theoretical analysis of the coarse to over-fine approach explains its advantages in handling flow-field discontinuities and simulations show its benefit for sub-pixel motion. By applying the suggested technique to various multiscale optical flow algorithms, we reduced the estimation error by 10%-30% on the common test sequences. Using the coarse to over-fine technique, we obtain optical flow estimation results that are currently the best for benchmark sequences.

Active Contours for Multi-region Image Segmentation with a Single Level Set Function

“…It should be also noted that the width of the -level sets influences the size of the smallest feature that the algorithm is able to segment. To capture small features one may up-sample the image before the segmentation, similar to the technique used in [2]. It is important to note the computational efficiency of the proposed method.…”

“…Comparison of the proposed method using the multi-region piecewise constant model, the convex relaxation approach of [7], and the graph-cut based method of [12]. [7] and (c), (d) [12], with different algorithm parameters This can be overcome by up-sampling the image prior to the segmentation [2]. It also should be noted that some of the object boundaries provided in the groundtruth segmentation and not detected by the proposed method, may be found only using a prior knowledge of the object structure.…”

Optical flow modeling and computation: A survey

“…It can be much more tractable to consider the expanded version of (3) with partial derivatives, resulting in a linear version of (4) the two-dimensional problem remains under-constrained. This is known as the aperture problem, stating that motion of linear structures, as it is assumed by (5), is by nature ambiguous if the neighboring context is not taken into account.…”

“…The idea is to create a pyramid of coarse-to-fine downsampled versions of the original image. At the coarsest level, the linearity domain of the image encompasses large displacements and the estimation can be based on (5). The estimations at coarser levels serves to warp the image at subsequent finer levels, where the estimation then reduces to a search for small motion increments.…”

Feasibility of controlling prosthetic hand using sonomyography signal in real time: Preliminary study

“…The process of extracting the SMG signal is similar to motion estimation or object tracking, which commonly make use of optical flow and BMAs [24][25]. BMAs are more popular because of their robustness in complicated environments with a low signal-to-noise ratio [24,26].…”