Fast and Dense 2D and 3D Displacement Field Estimation by a Highly Parallel Image Correlation Algorithm
Guy Le Besnerais, Yves Le Sant, David Lévêque · Strain · 2016
Abstract This paper describes a fast method for estimation of dense 2D and 3D displacement fields from image correlation. It is based on a previously published local, or window‐based, optical flow algorithm which is ideally suited for parallel processors. We describe the algorithm, its extension to stereo image correlation and its implementation on Graphical Processing Unit (GPU). We present the properties of the estimated displacement fields on simulated images and evaluate their accuracy on real data from a rigid body movement experiment. The main features of the method are a dense output (i.e. a 2D or 3D displacement vector per pixel) and a highly parallel structure which allows very high computational performance. A pair of 4 megapixels stereoscopic images is processed in less than 0.2 s. on a Titan GPU. Finally, we present and comment several experimental results obtained with the proposed method during mechanical experiments conducted at ONERA.