3-D object recognition from single perspective views

In this paper, a method to recover spatial information from single perspective 2-D views of 3-D planar faced objects is presented, which exploits the geometric constraints of central projection. Space junctions, that is the concurrences of three lines in a point in space, and angles have been considered as the main 3-D object descriptors. With respect to the approaches for determining the possible junction orientations reported in literature, analytical expressions to reduce the dimension of the solution space are derived, which can be useful to improve the 3- D object recognition process. A suitable database organization for efficient model retrieval is also proposed.