Application of neural networks in 3D object recognition system

The authors' objective is to create a system of automatic recognition of 3D objects. The authors propose the application of the projection moire method for acquisition of data about shapes of 3D objects. Neural nets play a crucial role in the system. Three basic parts of the system are presented in the paper. The first part is a data acquisition subsystem which contains an optic projection moire system and receiving module with the CCD camera and the frame grabber computer card. The second part is the feedforward neural network and other image processing software that pre-processes fringe pattern images obtained from the interferometry. The third part is the neural network realizing Self Organizing Map (SOM) clustering. The role of this net is automatic classification of particular parts of the image obtained at the second stage. Experiments confirmed that the proposed approach to recognition of 3D objects is very promising.