A Vector-Based Backward Projection Method for Robust Detection of Occlusions When Generating True Ortho Photos

Abstract

One of the problems in detecting occlusions when generating true ortho photos is false occlusion and false visibility caused by rolling terrain, and the resolution difference between the digital surface model (DSM) and the images. Published methods are plagued by unstable computation and have considerable noise. The paper presents a vector-based backward projection (VBP) method for detecting occlusions. First, the method projects vector polygons representing the surface of buildings to image space. Next, it derives their true coverage by judging the occupancy priority for the overlap. It then determines the visibility of building and digital elevation model (DEM) grids. Finally, it generates a true ortho image by differential rectification. In addition, this paper develops methods to keep all steps robust, such as dubious region growing methods for rasterization and judgment formulas for boundary noise elimination, which makes VBP practical. Experiments using the Z-Buffer method, the projection-ray-angle-based method and VBP with a typical image and DSM prove VBP has 100% detection accuracy, and is robust and efficient.