Multi-view re-projection for 3D-object reconstruction

H. El Hafed*, Z. Guennoun, A. Lichioui

L.E.C, Ecole Mohammadia d’ingénieurs Agdal, Rabat, Maroc

* Corresponding author. E-mail:,,

Received: 21 December 2006; revised version accepted: 01 July 2007



     This paper describes a robust approach for 3D-object reconstruction using parallel projection (orthographic views). For the 3D reconstruction, this method uses the projection of points of interest, instead of projecting every pixel (Voxel-map method). First, an isometric projection view is produced and obtained from a set of orthographic views: Front, Top, and Right Side views. As the modified Harris detector algorithm presents very high performances using few parameters, and not requiring any threshold values, it is used for extracting points of interest for each view. The reconstruction is obtained by re-projecting the points of interest that have just been extracted from each view. At the end, the 3D-object surface is rebuilt using the intersection of points of interest in the 3D space according to the appropriate direction. Finally, for smoothing purposes, a subdivision-based smooth representation is introduced that modifies the linear surface representation mesh.


Keywords:             Orthographic multi-view projections; Harris detector; 3D reconstruction; Surfaces subdivision.

© 2015