Optical flow in the presence of spatially-varying motion blur

Abstract

This paper extends the classical warping-based optical flow method to achieve accurate flow in the presence of spatially-varying motion blur. Our idea is to parameterize the appearance of each frame as a function of both the pixel motion and the motion-induced blur. We search for the flows that best match two consecutive frames, which amounts to finding the derivative of a blurred frame with respect to both the motion and the blur, where the blur itself is a function of the motion. We propose an efficient technique to calculate the derivatives using pre-filtering. Our technique avoids performing spatially-varying filtering (which can be computationally expensive) during the optimization iterations. In the end, our derivative calculation technique can be easily incorporated with classical flow code to handle video with non-uniform motion blur with little performance penalty. Our method is evaluated on both synthetic and real videos and outperforms conventional flow methods in the presence of motion blur.

Dense 3D Reconstruction from Severely Blurred Images Using a Single Moving Camera

“…(9) and then compare the motion vectors with the results of the blur-robust optical flow method in [19] to evaluate the pixel correspondence accuracy between images. As described in Section 2.2, two additional images are used in [19] as additional information, whereas camera motion is used in our method. The optical flow results are compared in Fig.…”

“…Estimation of motion blur kernels from two images utilizes the idea that applying the blur kernel of each image to the other image results in the same cumulatively blurred images [19,20]. Let I n−1 and I n be two consecutive blurred images in an observed sequence, which have latent unblurred images L n−1 and L n , as well as blur kernels K n−1 and K n , respectively.…”

“…Therefore, an additional constraint has to be incorporated to eliminate the ambiguities in v n−1 and v n . The ambiguities in motion or blur kernel estimation given two images has been addressed in several previous works [10,19,20]. For example, the directions of the blur kernels of two images are assumed to be known [20]; otherwise, additional input images are used to refine the motion vectors of the two blurry images [19].…”

“…1) enables handling of a curved motion path caused by camera rotation. Consequently, the proposed depth-based blur model can address more general motion blur compared with [19], where the blur kernel was assumed to be linear.…”

Blind motion deblurring using optical flow

“…In [30], a registration based deblurring approach has been proposed from two spatially variant blurred images. Some other works are to give flow estimation in the presence of motion blur [31][32][33]. They also used the commutativity of the blur operation to match blurred images, but did not deblur the image.…”