GENERATION OF FULL-SCALE 3D MODELS FROM 2D IMAGES PRODUCED BY A SINGLE-EYE IMAGING DEVICE

摘要

The invention relates to a method for producing a full-scale 3D reconstruction of a three-dimensional scene using a single-eye imaging device (11) having a phase (210) for training a CNN (60) comprising generating (211) pairs of learning data (50) for the CNN, with, firstly, a depth map generated from a 3D reconstruction of a three-dimensional scene obtained by multi-view stereoscopy on the basis of at least two 2D images of the scene by intrinsically and extrinsically calibrated image sensors, and, secondly, one of the 2D images of the three-dimensional scene. In a prediction phase (220), a full-scale 3D surface is generated of the scene of interest, and is reconstructed on the basis of an associated depth map (35) which is provided by the CNN from an acquired 2D image (25).