The invention relates to a method for modelling a scene from a plurality of maps representative of depth, each map representative of depth being associated with a view of the said scene according to a particular viewpoint. To optimize the fidelity and precision of the scene modelling, the method comprises the following steps: generating, for each map representative of depth, a surface representative of the said depth map in world space; estimating a volume bounding all surfaces generated, the said bounding volume being discretised into a plurality of voxels; associating attributes representative of transparency with the said voxels of the bounding volume; and generating a resulting surface according to the attributes representative of transparency associated with the said voxels. The invention also relates to a device for processing the data representative of the corresponding depth.