`Walking on two legs': alternate use of two representations of amatch during object recognition and location

摘要

Performance is presented for an object recognition and locationmethod which alternately optimizes two representations of the matchbetween object and image. The two representations are a geometrictransform mapping the model onto the image coordinate and a feature pairlist indicating the association of matched model/image features. Eachoptimization process has been designed to be history-free andconsequently the system is capable of correcting mistakes formed in anearlier phase when the matching was rough