A QUADTREE ORGANIZATION CONSTRUCTION AND SCHEDULING METHOD FOR URBAN 3 D MODEL BASED ON WEIGHT

The increasement of Urban 3D model precision and data quantity puts forward higher requirements for real-time rendering of digital city model. Improving the organization, management and scheduling of 3D model data in 3D digital city can improve the rendering effect and efficiency. This paper takes the complexity of urban models into account, proposes a Quadtree construction and scheduling rendering method for Urban 3D model based on weight. Divide Urban 3D model into different rendering weights according to certain rules, perform Quadtree construction and schedule rendering according to different rendering weights. Also proposed an algorithm for extracting bounding box extraction based on model drawing primitives to generate LOD model automatically. Using the algorithm proposed in this paper, developed a 3D urban planning&management software, the practice has showed the algorithm is efficient and feasible, the render frame rate of big scene and small scene are both stable at around 25 frames.


INTRODUCTION
With the rapid development of traditional mapping industry technology, the construction and application of 3D digital city become more widely, Has been developed to a city's 3D application from a single cell or small scene roaming browsing.The rapid development of geographic information acquisition technology, explosive growth of UAV technology, high-speed progress of Lidar technology, make Urban 3D model data acquisition methods become more and more diversified and convenient.Both the model accuracy and the model simulation have made great progress, also caused another problem that attract people's attention: The dramatic growth of data (DAI,2013;DENG,2013).Even the data a fine urban building model is likely to reach hundreds of megabytes.So even with the highly developed computer hardware, massive model data scene organization scheme and data scheduling strategy are still worthy of study.

QUADTREE INDEX ORGANIZATION FOR URBAN 3D MODEL BASED ON WEIGHT
Spatial data organization is a reasonable planning of space data and the establishment of spatial index, to improve the retrieval efficiency of spatial data (YAN,2001).The number of urban architectural models in the 3D digital city scene is huge, structure is complicated, it is unrealistic to load all the city models into the computer's Random Access memory in real-time scene rendering, it is necessary to query the eligible model data and load computer Random Access memory quickly.In order to complete the fast query of Urban 3D model, the spatial indexing of data is an indispensable step.Spatial Index of Urban 3D Model is a data structure arranged according to certain rules according to the location and shape of the Urban 3D model object.Contains summary information for spatial 3D models, such as model ID, minimum bounding box.The common spatial index including grid index, B-Tree, R-Tree, quadtree, octree and so on (Zhu,2006).Terrain, images and other spatial data organization mainly based on the grid division, but because of the irregularity of building model, grid division produces a lot of division at the grid boundary.B-Tree is a onedimensional index, can not process two-dimensional and multidimensional spatial data.R-Tree is natural expansion of B-Tree in multidimensional space, suitable for multi-dimensional space query, but due to the occasional spatial data distribution, the nodes are easy to overlap, will produce multiple branch inquiries in spatial queries, reduce the efficiency of space query largely, so it's not suitable for randomly distributed building models.
The quadtree is a kind of indexing mechanism that belongs to the spatial partitioning organization index structure (Pajarola R,1998), it's data storage structure is a hierarchical tree structure, which is widely used in 3D digital city because of the higher query efficiency.

EXPERIMENT ANALYSIS
Based on the study of this paper, as Xijing Hospital is building a hospital intelligent management system, we developed a set of ground and underground two three-dimensional linkage city planning and management system, the system utilizes 3D visualization technology on the basis of two -dimensional city information system, combine DOM, DEM, Fine 3D model, in order to complete intuitive, putative, interactive, all-round display of the status of urban construction, show the city real scene from the software by scientific means, provide decisionmaking functions for urban planning and municipal administration, reduce the loss caused by the timely interaction of information.
In order to verify the feasibility of the algorithm in this paper, we select two sets of data to do the test, a set of data is Xijing hospital test data which is relatively small, the other is Korla digital city data which volume is larger.As the picture shows.1)Xijing Hospital has a small amount of data, but the data contains terrestrial buildings and terrain data DEM,DOM and underground pipeline data.Hospital pipeline length is 105.428kilometers, the number of pipeline points is 12053.Collect texture data volume is 30GB, 5669 real photos.Establish 562 fine models.The efficiency of 3D applications can be measured by frame rate, which represents the number of times the system can draw every second while the scene is being processed.The higher the frame rate, the more smooth the rendering effect.In general, it is acceptable that the frame rate of three-dimensional application is 25 or so.As the algorithm design the system calculation is large when the scene moves, so the frame rate is slightly different when the scene moves and rests, we will test the frame rate when the scene move and static simultaneously.Some places of the 3D digital cities are densely populated, in some places the building is scarce, the rendering pressure of the two regions is quite different under the same conditions.This article will divided the region into three areas for testing: It can be maintained at 30 or more frame rate where the model is generally dense, basically there is no pressure to render.The gap begins to widen between the frame rate of still and moving where model is more intensive.It can be maintained at 24 or more in the still scene when both in large scenes and small scenes, when the scene is moving, the frame rate is reduced by 4,but still able to maintain more than 20 frames.Through the above test we can see that the proposed algorithm can be used to organize the 3D model of mass city, to keep the system to maintain more than 20 frame rate.

SUMMARY AND OUTLOOK
This paper proposes a Quadtree construction and scheduling rendering method for Urban 3D model based on weight, the method assigns weights according to the different characteristics of the city model, divide different levels by weight, and then the Quadtree is organized and dispatched.And divide the data into 3 layers of LOD can improve the scheduling efficiency effectively.This research achieved certain results, also developed a practical application system, but there are some areas for improvement, such as, the frame rate of moving scene is lower than still scene, the LOD model can generate more hierarchical models.With the oblique photography technology increasingly mature, the mixed rendering of tilting photography 3D data and urban 3D model is also a development trend, is the follow-up needs of this study.
Figure 1.Quadtree construction process3.SCHEDULING AND RENDERING METHOD FOR URBAN 3D MODEL

Figure
Figure 2. fine model

begin
Figure 5. the process of scheduling for Urban 3D model based on weight2)Korla data volume is large, the data covers an area of 300 square kilometers, Korla area is divided into two parts: the old city and the planning area, texture data more than 400GB,there are more than 800,000 urban 3D model data, contains more than 60000 fine models.The test machine is ordinary PC, the configuration parameters are: CPU: Intel(R)Core(TM)i5-3470, 4-core 3.20GHz,RAM: 8G, Video card: NVIDIA GeForce GTX 960, graphics memory: 2G, Operating system: WIN7 64-bit operating system.

Figure 6 .
Figure 6.The efficacy of Xijing hospital Data rendering

Table 1 .
Two frame rate statistics Model density Model dense area, Model medium area, Model sparse area.we divided the area into dense areas if each frame rendering more than 4 million, medium area if each frame rendering is 2 million to 3 million, sparse area if each frame rendering under 2 million.The final test results are shown in the table 1.The frame rate listed in the table is the average frame rate obtained by sampling 20 times for each region and state.

,
It is easier to render in a sparsely populated place, it can be maintained at 50 or more frame rate in any case.