Variational Optical Flow from Alternate Exposure Images

Abstract

Traditional optic flow algorithms rely on consecutive short-exposure images. In contrast, longexposed images contain integrated motion information directly in form of motion blur. In this paper, we show how the additional information provided by a long exposure image can be used to improve robustness and accuracy of motion field estimation. Recently, an image formation model was introduced [23] that relates a long-exposure image to preceding and succeeding short-exposure images in terms of dense 2D motion and occlusion. We formulate the original two-step problem for motion and occlusion timings as a joint minimization problem and derive a global TV-L1 energy functional that can be minimized efficiently and accurately. The approach is able to calculate highly accurate motion fields, assigning motion to occluded and disoccluded image regions in one joint optimization procedure.