A 3D shape identification system using a fiber grating vision sensor

The authors present a system for quickly identifying the shape of a 3D object using space encoding and a pattern transformation of image data. In this system, a dot-matrix-like pattern of bright spots is projected onto the object surface area, using the optical diffraction effect produced by a fiber grating device. A spot projection image, which is obtained from the vision sensor, is transformed into a pattern of invariance to the up-down and/or left-right translation of the object. The system can produce an output of invariance to the translation and the rotation of the object by training a multilayered neural network using the transformed pattern. Results of experiments performed to verify the effectiveness of the proposed method are discussed