Abstract: To improve the performance of a 3D object recognition system, the extraction of the invariant moments of 3D objects as object features, together with the modified BP neural network, is used for 3D objects classification and recognition. The theoretical analysis and simulation prove that using the invariant moments feature of 3D objects has the ability to make classification and recognition. The analysis of its is further principal components made to process these invariant moments features to get better recognition performance. A 100% classification rate can be obtained, and the complexity and training time of the neural network are reduced.

Key words: 3-D object recognition, invariant moments, neural network, principal components analysis