METHOD FOR DETECTING A TARGET IN STEREOSCOPIC IMAGES BY LEARNING AND STATISTICAL CLASSIFICATION ON THE BASIS OF A PROBABILITY LAW

摘要

The present invention relates to a method for detecting a target present in at least two images of a single scene acquired simultaneously by separate cameras. The method comprises, in focusing conditions, a preliminary step of learning targets, said learning step including a step of modeling data X corresponding to an area of interest in the images using a distribution law P such that P(X) = P(X_2d, X_3d, X_T ) = P(X_2d )P(X_3d )P(X_T ), where X_2d represents luminance data in the area of interest, X_3d represents depth data in the area of interest, and Χ_T represents movement data in the area of interest. The method also comprises, in operation conditions, a simultaneous step of classifying the objects present in the images, the target being considered as detected when an object is classified as being one of the targets learnt during the learning step. The invention can be used for monitoring, assistance and security from stereoscopic images.