A Method For Refining Digital Surface Models Based On Projected Images And Its Applications

At present,with the emergence of the concepts of digital city,intelligent city,digital twin,and ocean resource development strategies,the demand for highly accurate three-dimensional spatial information is increasingly urgent.For example,the refined three-dimensional city models and ortho-rectified images can be applied to urban planning,urban monitoring,urban tourism,transportation,disaster monitoring,and other high quality ocean related remote sensing data can be used for monitoring of key applications,such as offshore marine survey,coastal mapping,resource survey,marine dynamic supervision,marine emergency response,marine resources and environment monitoring.Detailed DSM is the key to making high-quality orthoimages.The effective technologies of obtaining digital surface model(DSM)mainly include Li DAR(Light Detection and Ranging)and digital photogrammetry.The accuracy of 3D data obtained using Li DAR usually is higher comparing to stereo photogrammetry but the cost of using Li DAR is higher too in contrast to stereo photogrammetry.With the advance of imaging and computing technology and advanced dense matching algorithms,more and more applications tend to use stereo photogrammetry.Various kinds of data errors are common in reconstructed DSM,especially from stereo photogrammetry.The reasons cause those errors could be: the lack of textures or repeated patterns on images which confuse the image matching algorithms,occlusions due to limited overlapping and view angles,moving objects on the ground especially in urban environments and so on.To remove those errors and artifacts is a challenging task.Currently,the practical solution is still manual editing and manipulating those affected DSM regions.Although some degrees of semi-automation can be employed the DSM refinement remains a labor consuming and expensive process.This study proposes a new method to refine an existing DSM.The proposed method utilizes the concepts of “projected images” and “projected locus” to efficiently calculate the coordinates between image-space and object-space and simplify image resampling during image matching.Once the relationship between projected images and DSM,several image matching techniques can be easily implemented into the refining method.The normalized correlation coefficient(NCC)matching technique and probability relaxation matching techniques are illustrated and performed in the presented experiments.A series of experiments are conducted to illustrate the effectiveness and efficiency of the proposed refining method.And an extension is made to take into account the two-media case to handle bathymetric information extraction in shallow water areas.Using multispectral inversion technique to retrieve bathymetric information becomes popular due to the image availability and affordability.The main problem for the multispectral inversion technique is how to obtain the reference bathymetric information during model parameters fitting.This study develops another innovative idea to apply the refining method first to obtain the bathymetric information,then feed the newly obtained bathymetric information into the multispectral inversion technique,therefore an integrated approach can be formed to improve the practical use of both the refining method and the multispectral inversion technique.Encouraging results are obtained in the shallow water trials around Taihu Lake.