Method of recognition of objects in the images and its application to the following objects in the sequences of images.

摘要

The method of recognition of objects in the images according to the invention comprises a learning phase during which the object to be recognized selected in an image of the learning is modelled, and a recognition phase during which the zone modeled is compared with the image of the same size located in the images to be treated. The method used for the characterization of the image zones object model to recognize or areas of images to be treated, a decomposition of the matrices of the luminance values in the singular values, according to a matrix of projection matrices and of a diagonal matrix the ordinate, the latter constituting a signature characteristic of the zone in question; for the recognition phase, the differences between the signature of the model and the signatures of the zones of the considered image are calculated and the image zones are selected from the smallest deviations in an iterative method which consists in carrying out a coarse on a large number of zones in a first time, then a sorting of the end on a number of more and more restricted areas. The calculations carried out at each step are re-used at the next step.