Image flow segmentation and estimation by constraint line clustering

Abstract

Image flow is the velocity field in the image plane caused by the motion of the observer, objects in the scene, or apparent motion. The image flow velocity field is an important intrinsic image and many algorithms that use the image flow velocity field have already been described. The image flow velocity field can contain discontinuities due to object occlusion in the scene. An algorithm that can estimate the image flow velocity field when there are discontinuities due to occlusion is described. Experimental results on a demanding synthetic test case and a real image are presented. Some error analysis that explains the performance of the algorithm is provided. The velocity field estimate can be improved using surface reconstruction between velocity field boundaries. A new surface approximation algorithm that does not blur sharp boundaries is presented.

Face recognition technology for law enforcement applications

“…Other significant papers that deal with segmentation, motion detection from optical flow and normal flow may be found in [117,22,97,32,126,138,106,132].…”

Measurements of granular flow dynamics with high speed digital images

“…Most of the current approaches for the measurement of image motion deal with 2-dimensional motions in the image plane and can be classified depending on the choice of a measurement: (1) use of brightness variations over space and time to measure instantaneous image velocities, Le., gradient-based techniques (Horn and Schunck 1981;Schunck 1989); (2) measurement of displacement of local image pattern or primitive image tokens between successive frames of a sequence, Le., correlation -based matching techniques (Burt et al 1982;Glazer et al 1983) and symbolic-token based matching techniques (Prager and &bib 1983); (3) measurement of the spatio-temporal energy of the image brightness function in a local area to determine image motion, i.e., spatio-temporal energy model (Adelson and Bergen 1985;Watson and Ahumada 1985;Heeger 1987;Daugman 1989;Tsao and Chen 1991); and (4) update of displacements based on gradient search, Le., recursive displacement estimation (Musmann et al 1985).…”

Segmentation and factorization-based motion and structure estimation for long image sequences

“…Most previous motion estimation methods do not address the problem of segmenting the image sequence into regions corresponding to rigid objects with different 3D motions. Many that do ( [12], [13], [14]) segment the image at 2D motion discontinuities. Adiv [15] and Bergen et al [16] improve on this by instead segmenting the image based on how well the image flow fits affine motion models.…”