Method of detecting the relative depth between objects of a scene from a pair of images which may emerge in different views.

摘要

The present invention concerns a method for detecting the relative depth of two neighboring regions in relative motion relative to THE OTHER TWO FROM DIFFERENT VIEWS TOMAS IMAGES OF A SCENE TO KNOW THE COMPOSITION OF THE SCENE. According to this invention, the procedure consists of: - A SPATIAL SEGMENTATION one of the images, called reference image, - a local motion estimation between the two images, WHOSE vectorfield RESULTING CALLED LOCAL MOTION - A DETERMINATION OF ERRORS OF ESTIMATES OF MOVEMENT ALONG THE BORDERS OF MOVEMENT - AN ANALYSIS OF FINDING FAULTS motion estimation ALONG THE BORDERS OF MOTION TO CONCLUDE THAT A rEGION IS IN FRONT OF A rEGION B WHEN THE ERROR OF ESTIMATED MOVEMENT ALONG THE BORDER IS LOCATED MAINLY B-side of the border.