In this paper, we show that four uncalibrated images are sufficient to
uniquely determine the 3D projective invariants of a set of six point
s in general position in space. An algorithm for computing the unique
solution is proposed. It computes by solving a set of linear equations
for its two linear independent solutions and is simpler than other al
gorithms which compute the three possible solutions of 3D projective i
nvariants by solving a set of nonlinear equations. However, a fourth i
mage is needed to get the linear unique solution. Finally, experimenta
l results have shown the feasibility of this algorithm. (C) 1997 Patte
rn Recognition Society.