In this paper, we present an algorithm which is a possible model of the human biriocular shape reconstruction. In this new model, the shape of an object can be re(-onstructed from two perspective views with as few as three points, without explicitly estimating the object's pose. Simulation experiments show that the algorithm provides reliable results in the presence of noise. The psychological plausibility of this algorithm has been tested in a psychophysical experiment on shape constancy. Results of this experiment contradict prior models of human shape perception and are consistent with some (but not all) properties of our new alg~rit~hm.

Date of this Version

November 1996