Stereo Matching Based On POS Observations

With the development of photogrammetry technology, new sensors andtechnologies are constantly introduced to the application of photogrammetry.Integrating the Differential Global Positioning System (DGPS) with an InertialMeasurement Unit (IMU) we can get the six exterior orientation (EO) parameters ofimages when exposed and so allows for direct image georeferencing, offering apossibility of replacing the conventional aerial triangulation (AT) in airbornesurveying/mapping, which has become an active research point in contemporaryphotogrammetry.Image matching is a key problem in digital photogrammetry. Automatic imagemeasuring is to identify and to measure conjugate points in two or more overlappingphotographs automatically. In digital photogrammetry, this process is done by imagematching. It is shown that image matching is inherently an ill-posed problem and thesolution is usually not only one, so additional assumptions and constraints have to beintroduced to make it well-posed. An simple way to make image matching well-posedis to restrict the space of possible solutions and provide good initial parallax.Since the position and orientation system (POS) can get the EO of images,considering the well-known colinearity equation, we can predict the location ofconjugate points in stereo image in a good result, then a better initial parallax can begotten. Thus, the limitation of traditional method in deciding initial parallax, such asgaining incorrect initial parallax, large searching region, etc, may be hopefullyimproved. As a result, not only the operator's interference in automatic imagemeasuring will be reduced but also the efficiency and reliability of image matchingwill be improved.In this paper, the individual modules of some existing matching algorithms aredescribed and analyzed in detail. The stress is imposed on the studies of the use ofPOS observations in automatic image measuring. Using POS data in predicting thelocation of conjugate point can ensure a high reliability and avoid the dependence ofimage condition in image matching. At the same time, in this paper a grayvalue-based fast image matching algorithm is developed and implemented, whichintroduces POS observation and is characterized in making block pyramid image atreal time, local maximum, continue revising the matched point in every hierachicalimage matching, and so on. It also make it easy in multi-image matching inmulti-overlay region, providing a good method for direct georeferencing solving.