Object Recognition from Range Images Using Superquadric Representations

Segmentation of range images using superquadric entities has been pointed out by a number of researchers as a powerful approach towards object recognition. Problems exist in finding an unbiased decision function for the assignment of a superquadric representation with an object prototype. This paper discusses current distance measures between recovered models and prototypes and presents a novel method for classification of uncertain superquadric representations using a maximum likelihood criterium and incorporating the range image characteristics of an active optical triangulation sensor. The main advantage of this probabilistic method is its inherent minimisation of the classification error rate. The approach is applied to recognition of electronic components for printed circuit board waste management.

• Type: Conference
• Year: 1996
• Where: MVA
• Authors: Erik R. van Dop, Paul P. L. Regtien
• Keywords: MVA 1996, MVA 2007, Range Image, Superquadric Representations, Unbiased Decision Function