3D object recognition

Abstract

This thesis implements a 3D object recognition system by using Matlab software to simulate two methods: Principal Component Analysis (PCA) and Artificial Neural Networks (ANN). One 3D object can be described with several images taken from different positions. In order to have efficient storage and fast process, the number of data dimensions should be reduced. PCA method constitutes a new data space created by some eigenvectors of covariance matrix of image set. The successful recognition rate of the PCA method is about 80% for 12 training images per object and over 90% for 24 training images per object. The ANN method uses the data preprocessed from PCA method to train an artificial neural network for each object. With ANN is combined with PCA, the recognition result is better than only PCA method.