摘要
针对传统HS(Horn&Schnuck)光流估计算法存在奇异值、不能保留光流场边界不连续性、不能应对运动目标间相互遮挡等问题,提出一种联合后置滤波器的分数阶光流模型。在该模型中,应用分数阶微分处理HS模型中的平滑项以保留光流场的边界不连续性;通过分析结构张量特征值的数据特征来寻找光流场边缘点,利用联合流场散度与像素点投影差分的方法来检测遮挡区域;采用一种结合MF(median filter)中值滤波器、WMF(weighted median filter)权值中值滤波器、BF(bilateral filter)双边滤波器的CPF(combined post filter)联合后置滤波器,它能够通过检测是否存在遮挡、图像亮度不连续性、图像运动不连续性,自适应地调节光流场的扩散过程,从而获得更加准确的光流场。实验证明,该算法能在图像中存在光照变化、遮挡、多运动目标等复杂情况下精确地估计出光流场,且算法计算成本低,能满足实时性要求。
A combined post-filtering method for fractional-order optical flow model was proposed for improving the traditional Horn & Schnuck( HS) optical flow model about the following problems: illumination change,preserving the discontinuity,multiple motion problems and solve the occlusion problems. A fractional-order smoothness constraint equation was used to replace the integer order of smoothness constraint of the original HS model to preserve the discontinuity of the edge of the optical flow field. Structure tensor was used to detect the edge of the optical flow field,the combined flow divergence and pixel projection difference method is used to detect occlusion. An integration of median filter,weighted median filter and bilateral filter was proposed to post-filter the optical flow field to get higher accuracy optical flow while maintain the low computation load. Extensive experiments demonstrate the superiority of our algorithm.
引文
1 Horn B K,Schunck B G. Determining optical flow[J]. Artificial Intelligence,1981,17(3):185-203
2 Papenberg N,Bruhn A,Brox T,et al. Highly accurate optic flow computation with theoretically justified warping[J]. International Journal of Computer Vision,2006,67(2):141-158
3 Black M,Anandan P. The robust estimation of multiple motions:Parametric and piecewise-smooth flow fields[J]. Computer Vision and Image Understanding,1996,63(1):75-104
4 Smith A,Lane R G. Determining optical flow using a differential method[J]. Image&Vision Computing,1997,15(1):11-22
5 Chen D L,Sheng H,Chen Y Q,et al. Fractional-order variational optical flow model for motion estimation[J]. Philosophical Transactions of the Royal Society of London A:Mathematical Physical and Engineering Sciences,2013,371(1990):976-986
6 Rabe C,Müller T,Wedel A,et al. Dense,robust,and accurate motion field estimation from stereo image sequences in real-time[C]//The Proceedings of the European Conference of Computer Vision-ECCV 2010. Berlin Heidelberg:Springer,2010:582-595
7 Sun D,Roth S,Black M J. Secrets of optical flow estimation and their principles[C]//The 23th IEEE Conference on Computer Vision and Pattern Recognition(CVPR). San Francisco:IEEE Compute Society,2010:2432-2439
8 Wei M,Liang L,Pang W M,et al. Tensor voting guided mesh denoising[J]. IEEE Transactions on Automation Science&Engineering,2017,14(2):931-945
9 Xiao J,Cheng H,Sawhney H,et al. Bilateral filtering-based optical flow estimation with occlusion detection[C]//The Proceedings of the European Conference on Computer Vision. Berlin Heidelberg:Springer,2006:211-224
10 Wei M,Liang L,Pang W M,et al. Tensor voting guided mesh denoising[J]. IEEE Transactions on Automation Science&Engineering,2017,14(2):931-945
11 Baker S,Scharstein D,Lewis J P,et al. A database and evaluation methodology for optical flow[J]. International Journal of Computer Vision,2011,92(1):14-31
12 Butler D J,Wulff J,Stanley G B,et al. A naturalistic open source movie for optical flow evaluation[C]//The Proceedings of the 12th European Conference on Computer Vision. Berlin Springer,2012:611-625