A Method Of Overlaying Typical Linear Symbols On 3D Terrain Model With Optimization On Visual Quality

In the two-dimensional (2D) Cartography and Geographic Information Systems (GIS), map symbol plays an important role in alternatively highlighting spatial objects and generalizing spatial phenomena. In the three-dimensional (3D) GIS, interactive virtual geographic environment which combines Digital Orthophoto Map (DOM) and Digital Elevation Model (DEM) covers more complicated geo-referenced context than flat map. Therefore, it is necessary to introduce 2D map symbols (roads, rivers, residential areas, etc.) into 3D GIS, especially in semantic generalization and visual highliting.However, overlaying vectors on 3D terrain surface is no longer as simple as drawing them on flat map, especially those linear features with more complex structures, colors and styles. Based on this, visual artifacts generated by overlaying methods are easy to emerge, then cause information asymmetry between map makers and readers. For example, lines are punctured into or suspended from terrain models; serrated line edges (aliasing); uneven period length distribution of complicated line styles; intermittence of those thin lines far from view point and low performance of real-time interaction. These situations can easily cause the information asymmetry mentioned above.Supported by the National Natural Science Foundation of China-"Research on Overlaying Typical 2D Vector Features onto 3D Terrain Surface Algorithms" (41371365), this study presents a method of rendering and optimizing vector symbols with an inverse pixel calculation in screen space. The proposed method realizes a precise attachment towards some linear features with typical complex styles on 3D terrain surface, and avoids most of visual artifacts generated by conventional methods. In the CPU preprocessing stage, we propose that period division should follow the true length fit on the surface. A spatial index is also built in this stage. In the GPU rendering stage, pixels are projected from screen space to the 2D vector plane. And decide their color in accordance with positional relationship between vectors and pixels. Apart from this, an anti-aliasing algorithm is proposed under this circumstance too. In particular, GPU parallel calculation was used to accelerate the whole process in the method. The main contents of this research are illustrated as follows:(1) During data pre-processing period, toward those lines with typical periodically changed styles, the research study on their reasonable length distribution on 3D undulating terrain surface, and the construction method of style geometry. The method of widen lines into polygons is also mentioned, including rounded caps and dark edges. The construction of spatial index is studied to improve the search efficiency.(2) At real-time rendering stage, the relationship between screen coordinates and vector space coordinates are worked out, and the irregular quadrilateral coverage of a pixel in vector plane is acquired. The discontinuous problem in the projection process is also discussed and the corresponding solution is implemented at this stage.(3) In the projection transformation process, the research add a pixel multi-sampling strategy. The pixels with no needs to be resampled are excluded with a classification rule. By converting the sampling space, positional deviation of sample points is avoided. At last, the research formulate a proper coloring method of every sample point.(4) In consideration of high similarity of pixel calculation and the grid feature of device screen, GPU parallel schema is adopted to implement the proposed algorithm. The procedure of how CPU and GPU collaborate is studied in the research. The most proper number and size of block are tested in order to achieve optimal performance.After testing, proposed method owns the following advantages:precisely attachment to the terrain; rational cycle distribution; aesthetic visual effects and no possibility to appear artifacts like suspension or aliasing. Attribute to GPU acceleration, the real-time performance of viewing large scale scene reaches the level of 3D interaction.