Although much research has been carried out on feature extractionfrom 3D object, validity and appropriateness of matching measures havenot been given enough consideration. In 3D object recognition, objectfeatures are often represented by directional data, and treatment ofsuch data needs different operation from that of the data on Euclideanspace. In this paper, the authors propose a new method to define amatching measure between objects, and show that this represents thesimilarity reasonably well with good robustness. Moreover they show thatthe reliability of the partially input data can be estimated as aby-product of this method
展开▼