Local/global scene flow estimation

Abstract

The scene flow describes the 3D motion of every point in a scene between two time steps. We present a novel method to estimate a dense scene flow using intensity and depth data. It is well known that local methods are more robust under noise while global techniques yield dense motion estimation. We combine local and global constraints to solve for the scene flow in a variational framework. An adaptive TV (Total Variation) regularization is used to preserve motion discontinuities. Besides, we constrain the motion using a set of 3D correspondences to deal with large displacements. In the experimentation our approach outperforms previous scene flow from intensity and depth methods in terms of accuracy.

Dense Semi-rigid Scene Flow Estimation from RGBD Images

“…We use Teddy images of the RGBD dataset [17]. Unlike the Middlebury dataset, [13]. Results of the proposed method are clearly more accurate than those of [13].…”

“…The work by Herbst [6] follows this idea, but as [16], it lacks a coupling between optical and range flows, and the regularization is done on the optical flow rather than on the scene flow. In [13], a variational extension of [12] is presented. A weighted TV is applied on each component of the 3D motion field, aiming to preserve motion discontinuities along depth edges.…”

“…The Middlebury stereo dataset [15] is commonly used as benchmark to compare scene flow methods [1,8,5,13]. Using images of one of these datasets is equivalent to a fixed camera observing an object moving in X direction.…”

“…Using images of one of these datasets is equivalent to a fixed camera observing an object moving in X direction. As in [5,13], we take images 2 and 6 as the first and second RGBD image, respectively, and use the ground truth disparity map of each image as depth channel. Stereobased methods [1,8] do not assume RGBD images and simultaneously estimate the optical flow and disparity maps by considering images 2, 4, 6 and 8 of each dataset.…”

Efficient Dense Rigid-Body Motion Segmentation and Estimation in RGB-D Video

“…Hadfield and Bowden [2014] propose a particle-based framework to scene flow. Quiroga et al [2013] and Herbst et al [2013] pose scene flow estimation for RGB-D images in a variational framework. In [Hornacek et al, 2014], local point sets within sphere neighborhoods at each pixel are aligned in a randomized multi-step process.…”