Recognition of a solid shape from its single perspective image obtained by a calibrated camera

摘要

A new model-based method for object shape recognition is presented. This method consists in solving a set of linear equations and thus is computationally simple. We show how it leads to a hierarchy of geometrical representations of an object, involving similarity, affine and projective properties. Results of our tests with synthetic images illustrate that this method works reliably in the presence of large amount of image noise. Finally, we show that the method allows for speeding up the memory search for the correct model, which makes it more efficient as compared to conventional model-based methods.