Three-dimensional object classification using shadow moire´ and neural network

Abstract

A method of recognizing and classifying 3-D shapes with continuous surfaces by integrating shadow moire technique and neural network is presented. Unlike existing methods of 3-D shape recognition that use range images of polyhedral objects or objects of different geometries such as cones, rods, spheres, etc., the proposed method classifies continuous surfaces that are geometrically similar. The objects selected to test the classification method are eggs of four different grades. The shadow moire technique, which has greater sensitivity compared to structured lighting or laser scanning, is used to obtain moire patterns on the surface of the eggs. From the moire pattern images 14 parameters are extracted and used as input to a multilayer feedforward neural network. The results of the classification using the neural network show that the prediction accuracy attainable is 60% when classification is performed on all four grades. The accuracy increased to 95% when three of the grades are classified.