3-DIMENSIONAL OBJECT RECOGNITION USING SPHERICAL CORRELATION

Most research in 3-D object recognition has concentrated on feature extraction and paid much less attention to the computation and evaluation of matching metric for feature comparison. Many existing algorithms for object recognition have problems with noisy data and incomplete (partial) input data. This paper proposes a similarity metric based on extensive exploitation of 3-D features and characteristics of objects to solve the shortcomings of existing algorithms. Simulation and experimental results indicate that the proposed matching metric ranks the similarities among objects consistent with human intuition. It also is robust to noise and can even predict the outcome for a given noise level. Furthermore, in occlusion cases, the proposed algorithm is capable of recognizing objects using partial (incomplete) input data. It can also be used to evaluate the reliability and the contribution of a subset of the data relative to the overall object recognition task.