Jinyun Fang

A distributed geospatial data storage and processing framework for large-scale WebGIS

With the rapid growth of geospatial data and concurrent users, the state-of-the-art WebGIS cannot support massive data storage and processing due to poor scalability of underlying centralized systems (e.g., native file systems and SDBMS). In this paper, we propose a novel distributed geospatial data storage and processing framework for large-scale WebGIS. Our proposal contains three significant characteristics. Firstly, a scalable cloud-based architecture is designed to provide elastic storage and computation resources of shared-nothing commodity cluster for WebGIS. Secondly, we present efficient geospatial data placement and geospatial data access refinement schemes to improve I/O efficiency. Thirdly, we propose MapReduce based localized geospatial computing model for parallel processing of massive geospatial data, which improves geospatial computation performance. We have implemented a prototype named VegaCI on top of the emerging Hadoop cloud platform. Comprehensive experiments demonstrate that our proposal is efficient and applicable in practical large-scale WebGIS.

Fast map projection on CUDA

Map projection is a key task in cartography that transforms the geographical coordinates from one coordinate system to another. It has been widely used in the Geographic Information System application. However, map projection is a very time-consuming task, and fast processing speed is often required in interactive GIS scenarios. Parallel computation provides an opportunity to reduce run times. Nowadays, with the Common Unified Device Architecture (CUDA) becoming a new programming approach of the general purpose computing on graphics processing units (GPGPU), map projection can be accelerated by fully utilizing the computing power of this revolutionary technology. In this paper, we present an innovative implementation of this compute-intensive operation. We also explain the parallel characteristics of the graphics processing units (GPU) through the work assignment. The result shows our optimized implementation can obtain at least 6× to 8× speedups over a very similar single-threaded CPU implementation.

A transit path planning model based on the heterogeneous road network

Optimal path planning is the critical technology for public transit system. In this paper, we propose a new transit path planning model— 3-layers heterogeneous road network model (3-LHRN). Based on this model, we can generate optimal path results, making a good match with real life. Some approaches have been explored in the past to tackle the optimal path planning issues. Taking aim at the least transfers, the approaches are based on transit matrices of the map. However, the precondition of the algorithms is that transfer stops must be the same, and it is required that the origin and destination points must be the exact bus stops. These preconditions cause some inconveniences and inaccuracy. For instance, the final path results may have more than two transfers without acceptable walking distance. Besides, even though passengers know some marking buildings around the bus stops, without names of the particular stops, they cannot inquiry the optimal path. With respect to the model based on heterogeneous road network, we address the following two key technical issues: (1) setting up the heterogeneous road network with three layers: public transit layer, road layer and POI (Point of Interest) layer. The three layers form a unified entity by coordinates and attributions in order to provide enough information for the optimal path. They communicate with each other by exchanging spatial data and attribute data, (2) computing the optimal path according to the heterogeneous road network. We get appropriate bus routes from public transit layer, locate the desired positions close to some particular bus stops according to POI layer, and compute the walking distance between POI points and bus stops. At last, we generate the final optimal path from the information computed as above. The transit path planning model based on heterogeneous road network is applied to the public transit network of Beijing, China. It is demonstrated that 3-LHRN model can produce much more optimal paths with fewer transfers than ordinary models.

A new R-tree node splitting algorithm using MBR partition policy

In this paper we introduced a new R-tree node splitting algorithm. As an indexing technique for multi-dimensional data, R-tree is widely used in geographical information systems, CAD systems and spatial databases. An R-tree consists of nodes which in turn consist of records. Each node in R-tree must contain limited number of records in order that it can be stored within one disk block, thus a node splitting algorithm is used while inserting a new record into a full node. A node splitting algorithm is one of the crucial factors of the query performance of an R-tree since bad splits would result in an inefficient R-tree structure. At first we gave an efficient, linear node splitting algorithm that could construct an R-tree fast, by partitioning the minimum bounding rectangle of the node according to the shape of the nodes MBR and the shapes of the MBRs of the nodes records. Then we found that this node splitting algorithm would generate uneven nodes sometimes, that is the node to be split might be split into two nodes with one of them containing less records than the minimum number of records required. We then developed an algorithm to balance those uneven splitting results to meet the demands of the R-tree definition. At Last we improved our node splitting algorithm by considering the siblings of the splitting node. The siblings of the splitting node were picked up during the phase of choosing node to insert record and were put together with the splitting node to generate a better result. We performed several experiments to compare our node splitting algorithms with some other node splitting algorithms on both synthetic and real world data. These experiments tested the tree-construction costs and the query performance of the resulting R-trees. The results showed that the tree-constructing cost of our algorithm was lower than others and although the balancing procedure degraded the performance, our algorithm outperformed other node splitting algorithms in queries of the resulting R-trees.

An accelerated approach to create 3D Web cities

Creating 3D city from the geographical information is an important goal in the field of visualization for GIS and the needs for Web GIS are growing and expanding rapidly. Many scenes of the 3D cities are built based on the models which are made by computer aided designing software to present real scenes, however, this method costs too much time and labor with the low efficiency. This paper presents a method that can promptly construct a 3D city on the Web according to the geographical information. Our approach uses the descriptions for the city from the geographical and additional information to construct the scene and divides the buildings into “molecule parts”. With the bottom-up process from cursory objects to fine buildings to assemble the whole scene, the patterns correspond to Web environment which needs the distributed data management and gradual transmission. Ever divided part which can be classified into the common types and saved into distributed storage environment is constructed according to its characteristics. In addition, we first establish abstract databases to store and classify various “molecule parts” which are created based on geometrical arithmetic, common objects and small models. The descriptive information specifies the data from database, which just need loading rather than producing, with lower cost of time and labor. Our approach designs a processing assembling method to match the different parts of building and the level of matching can be elaborated further with the expanding interfaces. Our approach shares and reuses the resources for rendering, such as geometrical structures, materials, textures and even the small models, to diminish the amount of data and the influence of network delay. Experimental results show that our approach increases the speed for rendering and decreases the transmission capacity greatly. Our approach is appropriate for the Web environment to improve the rendering efficiency and for promptly creating a 3D city.

Research on vector spatial data access based on Main memory database

This paper concentrates on reducing the cost of I/O and improving algorithm efficiency by designing and realizing a spatial data access system based on main memory database (MMDB). We use Redis to manage vector spatial data in main memory which is faster than hard disk or disk-based database in respect of data access. We design a management strategy of spatial data, internal and external data exchange strategy and data backup mechanism. Based on MMDB, we design the model of vector spatial data to describe its form, organization and access methods. The experimental results show that our data access strategy can cut down the time of I/O and improve the algorithm efficiency effectively.

A parallel strategy for plane sweep algorithm in multi-core system

This paper proposes to design and realize a parallel strategy based on plane sweep algorithm that can be used in the multi-core system. The main contribution of this work is to present an efficient method to compute the large and complex data with plane sweep method. We use MPI (Message Passing Interface) as the programming model in Linux multi-core system, select shape file as the standard input and output files, and present a dynamic tree merge model to collect the result from each computing core, in which the time complexity can be reduced from O(n2) to O(n log n) and the problem of data lock and load balancing can be effectively solved. Experiment shows that in single-core system, our algorithm has a better performance than ArcGIS; in multi-core system of 8-cores, our algorithm can obtain 8× speedups compared with ArcGIS.

VegaMinerPOI: A spatial data mining system for POI datasets

During the past few years Special-purpose data mining systems have drawn great attraction in research and industry area for their application into real environments. In this paper, a new scalable spatial data mining architecture for POI(Points of Interest) datasets has been proposed on VegaGIS 3.0 platform to design and develop a professional spatial data mining system. Our research covers a wide spectrum of knowledge discovery, including: knowledge discovery and data mining process model, interface standards and language standards. The interface and architecture of some commercial spatial data mining system such as Oracle Data Mining (ODM) are also examined.

A parallel algorithm for line segment intersection

The algorithm of line segment intersection is the core algorithm for vector map overlay. The traditional method is based on the plane-sweep process to find the intersections of two groups of lines. Although such method proves to be efficient in serial system, the basic data structure of this kind of algorithm needs to maintain a tree structure, which has the data-lock in multi-core system. In addition, with the development of data volume and the enhancement of user requirement, a parallel algorithm is needed in order to compute the line segment intersections correctly and efficiently. In this paper, we present a parallel algorithm for line segment intersection to process the large and complex spatial data, which is efficient in multi-core system and cluster environment. In the experiment results, our algorithm leads to an accurate result compared with ArcGIS. And in the test of large spatial data, our algorithm demonstrates a good performance of speed-up with the increasing of cores.

One rasterization approach algorithm for high performance map overlay

It is a critical technique in GIS about how to implement map overlay with high performance. To avoid the disadvantages of computational geometry algorithm approach, an optimal algorithm based on rasterization is proposed in this paper, which adopts approach of rasterization of vector polygons with run length encoded scan line, map overlay on the basis of raster representation and vectorization of binary raster data with improved boundary tracing method. Experiments show the significant performance and efficiency improvements of the proposed algorithm, which is one promising solution for the map overlay function of on-line map service system taking the approach of pre-rendered raster tiles with predefined map scales.