Hybrid Correlation Algorithms. A Bridge Between Feature Matching and Image Correlation

摘要

The map matching problem has been in search of an 'optimal universal' matching algorithm since its inception. Because of difficulty in (1) defining a performance criteria for both accuracy and probability of correct match, and (2) in knowing a priori the distributions associated with all map errors, most researchers have resorted to the use of 'ad hoc' algorithms. These have generally been divided into two classes--feature matching and correlation. Up to the present time there have been two basic classes of map matching algorithms--those based on feature matching techniques and those based on image correlation. This paper describes a new class of hybrid correlation algorithms which incorporate features as an integral part of the matching process. These algorithms can be implemented such that it is not necessary to extract features from the sensed image. This paper concludes by showing the domains in which each class of matching algorithm (feature matching, image correlation, and hybrid algorithm) is most appropriate.