IMPLEMENTING SIFT AND BI-TRIANGULAR PLANE TRANSFORMATION FOR INTEGRATING DIGITAL TERRAIN MODELS
Since their inception in the middle of the twentieth century, Digital Terrain Models (DTMs) have played an important role in many fields and applications that are used by geospatial professionals, ranging from commercial companies to government agencies. Thus, both the scientific community and the industry have introduced many methods and technologies for DTM generation and data handling. These resulted in a high volume and variety of DTM databases, each having different coverage and data-characteristics, such as accuracy, resolution, level-of-detail – amongst others. These various factors can cause a dilemma for scientists, mappers, and engineers that now have to choose a DTM to work with, let alone if several of these representations exist for a specified area. Traditionally, researchers tackled this problem by using only one DTM (e.g., the most accurate or detailed one), and only rarely tried to implement data fusion approaches, combining several DTMs into one cohesive unit. Although to some extent this was successful in reducing errors and improving the overall integrated DTM accuracy, two prominent problems are still scarcely addressed. The first is that the horizontal datum distortions and discrepancies between the DTMs are mostly ignored, with only the height dimension taken into account, even though in most cases these are evident. The second is that most approaches operate on a global scale, and thus do not address the more localized variations and discrepancies that are presented in the different DTMs. Both problems affect the resulting integrated DTM quality, which retains these unresolved distortions and discrepancies, resulting in a representation that is to some extent inferior and ambiguous. In order to tackle this, we propose an image based fusion approach: using the SIFT algorithm for matching and registration of the different representations, alongside localized morphing. Implementing the proposed approach and algorithms on various DTMs, the results are promising, with the capacity correctly geospatially align the DTMs, thus reducing the mean height difference variance between the databases to close to zero, as well as reducing the standard deviation between them by more than 30 %.