Jara J, Cerda M, Delpiano J, and S Härtel
Image Processing On Line, 5 (2015), pp. 139�158.
ABSTRACT
The optical ow (OF) approaches for motion estimation calculate vector elds which determine the apparent velocities of objects in time varying image sequences. OF was introduced in 1981 by Horn and Schunck (HS) as 2…the distribution of apparent velocities of movement of brightness patterns in an image” [1]. The idea contains two basic assumptions: “brightness value constancy” and “smooth ow of the intensity values” between two successive images. At about the same time, Lucas and Kanade (LK) presented a method for motion estimation between images, considering constant motion patterns for image patches [2]. While the HS model is intended to solve the problem for a non-null motion eld over the entire image -a global approach- the LK model can produce homogeneus piece-wise motion eld “patches” -as a local approach- allowing null motion regions. Several variations of the original HS- and LK-OF approaches have been published. The presented algorithm is known as the combined local-global (CLG) method of Bruhn et al. [3], encompassing properties of both HS- and LK-OF models. CLG-OF aims to improve the accuracy of the OF motion eld for small-scale variations while retaining the HS-OF bene ts of dense and smooth vector elds.
An online implementation of this algorithm is available for 2D images, using a numerical scheme proposed by Bruhn et al. which is suitable for multi-grid computation [4]. It must be noted that the algorithm works on grey-scale images. Color images will be converted to single channel (grey-scale) prior to the online OF computation.