Simulation of Two Forms of Eye Motion and Its Possible Implication for the Automatic Recognition of Three-Dimensional Objects

Simulation of two forms of eye motion was undertaken as a step toward knitting together the visual system of a robot being designed to automatically recognize irregular three-dimensional (3-D) objects in any orientation. Included in this visual system are a binocular graded-resolution video camera assembly (BGRVCA), a planner, and a self-command computer. Noton and Stark hypothesized, first, that the representation of an object in memory is a piecemeal affair and, second, that both these pieces and the eye motions required to view the features they represent are linked together in what they call a "feature ring," indicating the sequence in which the features usually are examined. Simulation was performed of both the generation of such a sequence and the subsequent adjustment of fixation reported by Richards and Kaufman. An attempt was made to describe the extension of the two Noton and Stark hypotheses to apply to 3-D objects. The work of Oshima and Shirai in automatically recognizing planar and curved 3-D objects was presented as a partial fulfillment of the first extended hypothesis. The work of Marr and his associates was presented as a possible fulfillment of both extended hypotheses.