A framework for generating high quality digital elevation models in urban areas

Digital elevation models (DEMs) are very important for a variety of applications, such cartographic analysis, mobile communication, urban planning, and visualization. A conventional stereo pair is not sufficient to generate a complete and accurate digital elevation model (DEM) particularly in urban areas. Hence, a robust technique for DEMs generation is vital. This paper presents a technique that can match a large number of overlapping images. The traditional cross correlation technique is used to generate a good approximation of the DEM. Least squares matching is then used to generate a higher accuracy DEM. Due to discontinuities that result from man made features, the generated DEM still lack the complete surface representation. A refinement process is then applied to the generated DEM. The refinement process utilizes the geometrical properties of man made features such as buildings. Points contributing to buildings are delineated and used to fit planes in the object space. A robust estimation technique is used in this step to guarantee elimination of outliers. The final results after this refinement step looks appealing and suggest the use of this algorithm for surface reconstruction.