DEM Refinement Based On Shape From Shading And Its Evaluation

Digital elevation model plays an important role in the fields of civil engineering,earth science,resource planning and management,surveying and mapping,remote sensing,military affairs and so on.However,with the development of the times,these areas put forward higher requirements for DEM-higher resolution and precision.The commonly used DEM production methods,such as manual measurement,image matching,laser detection and measurement or synthetic aperture radar,often need to spend a lot of time and labor costs.The DEM interpolation method is relatively simple,but its accuracy is easy to be affected.Therefore,it is necessary to study a new method of DEM refinement.The Shape from Shading method can use a single image to reconstruct the shape of the object,which provides a theoretical basis for the further refinement of DEM by using the remote sensing satellite image to reconstruct the shape of earth.At present,several international institutions provide free DEM data covering the world(such as the above 90 meters SRTM DEM and 30 meters ASTER GDEM and AW3D30).At the same time,a number of medium resolution satellite image data(such as 30/15 meters Landsat 8 image,16 meters GF-1 wide image)also provides for free.These data have a shorter return period and can quickly complete the global coverage,which provides a data base for using remote sensing satellite image to refine DEM.This paper studies the DEM refinement method based on SfS theory.Based on the principle of SfS and combined with the specific characteristics of remote sensing satellite images,we refine the DEM through the establishment and solution of physical imaging model.First of all,in order to eliminate the influence of atmosphere in the imaging process of remote sensing satellite image,we modeled the radiative transfer and use the least square method to solve the atmospheric parameters.And then,in order to explore the influence of the surface reflection model on the refined results,the Lambertian reflection model and the Lommel-Seeliger model in non Lambertian reflection model are respectively used for experiments.Finally,in order to find out the influence of the solution method on the refined result,the minimization method and the linearization like method are used respectively.This paper takes Hubei Province as the experimental area,using GF-1 wide image and SRTM as the experimental data.The accuracy of the results is evaluated by the detail,information and precision.The results show that the DEM refinement method based on SfS theory is better than the three order spline interpolation method in terms of detail richness,information content and accuracy;The refined result of DEM in the non Lambertian reflection model is nearly equal to the Lambertian reflection model in detail,information content and precision;The results of DEM refinement under the linearization like methods are better than the minimization methods in detail richness,information content and accuracy.