QUADRICS-BASED MATCHING TECHNIQUE FOR 3D OBJECT RECOGNITION

We propose a new method for the recognition of objects from range data by matching the features of the observed surfaces in the scene with the features of the model surfaces. The surfaces are represented as quadrics in which planes are treated as a special case. The parameters of quadrics are converted into features which include Euler parameters accounting for the orientation of the surface. The matching process consists of finding a transformation of the model surfaces and devising a measure of consistency, such that the mean square error between the feature space is a minimum. The matching involves a tree search procedure with backtracking whenever error exceeds the limit.