ABSTRACT
Embedded GIS is an embedded system production integrated GIS function, which can be considered as a GIS running on embedded equipments or an application of embedded system in spatial information. Compared with desktop system, the embedded system possesses limited processing capability and memory capacity and smaller screen and resolution. To implement GIS function in embedded system, it's necessary to fully exploit the efficiency of data model and look for an excellent model, making up the disadvantages of hardware in this platform and achieving fast graphic display and highly efficient search. The map's details expressed or concerned should not stay same under different scales. Those details showed under large-scale are not always displayed under small-scale. There are two approaches to meet this demand. One is to dynamically determine which data to display or not according to current scale and prearranged display rule. The other is to display an appropriate edition of map suitable for current scale, which is chosen from ready-made editions under different scales. The former is an approach to sacrifice time to gain space which demands powerful processing capabilities. The latter is to sacrifice space to gain time which requires large memory capacity. Processing capability and memory capacity in embedded system are both limited. Adopting either of the two approaches directly won't produce an effective result. The two approaches should be integrated to find a load balancing solution on time and space to reduce read-write operations and decrease storage occupation, so as to satisfy the needs of fast graphic display and highly efficient search. This article presents Object-Oriented Discrete LOD to solve the problem of displaying map's hierarchical details in embedded system. From the viewpoint of map expression, the map's details or contents expressed under different scale may not be same. Those details appeared under large-scale are not always displayed under small-scale. If excessive details were displayed in the map under a certain small scale, the efficiency of display would decrease and the key points expressed in the map would be confusing. This disobeyed the principle of integrative-mapping in cartography as there are conflicts among numerous details (Wu Lun et al., 2001). Maps under different scales should express details or contents at different levels. When map's scale is changing from small scale to large scale, richness of displayed details may be increased so that map data under different scale can't remain the same. The desktop GIS system is generally taking a uniform processing method when the map scale is changing, which only uses one map version containing whole data and only changes the display scale rather than the map detail. This processing mode simplifies the organization and display of map data but demands powerful processing capabilities. With better processing performance, it is adopted abroad in desktop GIS systems. However, the embedded system possesses limited processing performance. Using one map edition containing all details under any scales will make high consumption of processing resource and lead to a slow system response of map browse operations such as zoom out and zoom in etc, especially under a smaller scale. In addition, the screen of embedded system is small so making all the details gathered under a smaller scale will cover the key points instead. The LOD Model, which also named as Multiresolution Modeling, can achieve the goal of displaying different hierarchical details of a map according to scale. It is a set of models memorizing scene or different details of objects in scene and one detail level is one model (Guo Yangming et al., 2005). The fundamental is that as optical spot is getting closer to object, more details can be observed. According
to certain judgment about distance, the mapping procedure displays corresponding details appropriately. At present, the LOD model is most commonly used to create vivid and smooth scene in 3D graphic simulation environment and virtual reality (Zhang Guoxuan and Wei Hui, 2001). There is a correlation between scales of map and detail levels in LOD. The displayed details get richer when map zoom in from small-scale to large-scale, Just like the shift from low-resolution detail level to high detail level. High consumption of processing resource can be avoided while not displaying those insignificant details by the means of dividing a map into different hierarchical details according to scale, which also make the plot more clear and tidy and easier to be distinguished, so the interested areas or attended sites can be located more quickly and efficiently. Common LOD Model includes Discrete LOD and Continuous LOD (Zeng jun and Zhang Yuepeng, 2000). The Discrete LOD stores different levels of details as separate files. It has the following advantages: pre-created individual levels of details, avoiding consumption of system resource while constructing level of detail at run time, preknown relations and simplification among levels, simple calculation and easy judgment on the critical condition of Model. Critical condition is the switch condition in the Model from one level of detail to the other. The Discrete LOD also has disadvantages, like that the levels of details are predefined and can't be modified by user, and that switching between levels can introduce visual pops effect. Interpolating between levels can smooth the graph, but it causes new consumption of processing. The Continuous LOD is the Model that dynamically creates models at run time according to critical condition and certain arithmetic. It can construct any desired level of detail theoretically and thus is called Continuous Model. Compared with the Discrete Model, it has the following advantages: needs less storage spaces by only saving the highest level of detail, more levels of details are available which can be specified by users, adjust detail gradually and incrementally without visual pops. But it needs additional consumptions of processing resource and time. The LOD can be introduced to solve the problem of displaying different hierarchical details of a map in the embedded GIS that using the scale as critical condition. In this way, the critical condition is a certain predefined scale, for example, 1:500, 000. So when the current scale is smaller than 1:500,000, the map containing corresponding details will be displayed. As the current scale is larger than 1:500,000, more details will be displayed. This LOD using the scale as critical condition also includes Discrete LOD and Continuous LOD. The map was organized as editions of different levels of details when the Discrete LOD applied with certain scales as critical condition. Only the edition of highest level of detail was saved permanently from which temporal edition of other levels can be extracted dynamically when the Continuous LOD applied. There is only one permanently saving edition which is also the one containing all the details and other temporal editions were created based on this edition according to current scale. The ways for displaying the map under different scales as the Continuous Model and desktop GIS process Model applied are still different. Although only one edition was preserved permanently in both Models, temporal editions will be constructed according to the current scale when applying the Continuous Model; however desktop GIS process Model won't do it. It has been discussed that the Discrete Model and Continuous Model use scale as critical condition. The Continuous Model needs much higher proceeding capability than the Discrete Model does. But it demands much more storage spaces for map's editions when using the Discrete Model. And it will be hard to maintain uniformity among different editions. Those problems will become obvious if more detail levels were taken. Because of limitations of processor and memory on embedded platform, the two models can't be applied to the embedded GIS directly. There is a kind of inclusive relationship among map's editions when using the Discrete Model. Highlevel edition in hierarchy includes all details that low-level edition holds, and some extra details that low-level edition does not hold. According to thoughts of Object-Oriented, editions with different hierarchical details can be regarded as classes, and the inclusive relationship among editions can be structured as inheritance relation between superclass and childclass. High-level edition in hierarchy is derived from low-level edition, with some extra details. Two editions share all details preserved in lowlevel edition. Only extra detail data should be saved in high-level edition. They also share the map operations as zooming, panning etc. So this inheritance relation is formed among editions with different hierarchical details. The share among editions can be helpful on maintain the consistence of different details. The shift from low-level edition to high-level edition is due to the changes from small scale to large scale. Scale is the critical condition which leads switch of detail level. Therefore, this inheritance relation can also be considered as inheritance of map's editions under different scale. In virtue of this inheritance relation, the same data of map's editions could be shared. Edition under large-scale is derived from the one under small-scale, only saving the detail data of its own. The same
data is only stored in edition under small-scale, not under large-scale. Thus, it can save much memory space, and will be helpful to keep identical among those editions. When map updating is taken up, the modification can be restricted within the edition which stores details in change. So consumptions of system resources can be reduced, and identicalness can be ensured among all editions that share changed details. We carried out experiments of handling hierarchical details, using four models and a part of a country cadastre map. Effect comparison of those four models will be done. The experiment material is a 20kmx20km covered cadastral map of Gong An Country, with 387 points (annotations) and 125 polygons, the size is 3,073KB. HP iPAQ H5550 pocket pc is chosen as research platform with the hardware components as: Intel Xscale PXA 255 processor at 400MHz, 128MB RAM, 48MB ROM, a LCD of 320 x 240 pixels and the operating system as Microsoft Pocket PC 2003. Microsoft Embedded Visual C++4.0TM is chosen as software development environment. The cadastre map is asked to be divided into three levels of details and the scales as critical conditions are defined as 1:200000 and l:50000(not required in the desktop GIS process Model). Experiments were carried out with the four specific Models and each experiment consisted of two steps. The first step is data organizing and development of map browser function and the second is testing and recording. In order to simplify the development task, map browse function only includes four sub-functions: data loading, zooming out, zooming in and panning zooming. Commonly, those operations are high frequency and basically operations when implement a GIS function. The testing includes four items with the purpose of investigating on the consumptions of CPU processing resource and storage capacity using the four Models. We get the average values of 30 times randomly recordings of those consumptions. The traditional processing method of Desktop GIS process Model is the frame of reference and comparison among the other three models which can give different levels of detail under the different scale. Each dynamic details model is compared with Desktop GIS process Model to give the relative radio. The Continuous LOD occupies the largest computing time as 5.7 seconds and the smallest space as 3.0 MB, equal the original size of the data, because it draw the different models at real time with compliant to the scale. The Discrete LOD reduce the processing time to the lowest one of 2.7 seconds through the way of prepare the edition's copies in advance. Our Object-Oriented Discrete LOD is a load balance strategy that it depress the amount of storage consumption, and it is economize the consumption on the processing possess at the same time. Consider that Desktop GIS process Model can't provide the changed details under different scale, it will be large waste on the limit processing and storage space in the embedded system. With the sacrifice of extra approximately 10% consumption on the storage space, Object-Oriented Discrete LOD give a more faster response on usual map operations and provided the similar visual effect. The consumptions of CPU and storage using the four Models are generally consistent with the theoretical prediction. However, the Continuous model needs the most storage spaces. Possibly it is associated with that larger memory spaces are demanded when creating temporary copies. According to effect comparison of those four models, we concluded that the requirement on processing resource and storage capacity while using the Object-Oriented Discrete LOD is moderate. The aim, that displays different hierarchical details of a map according to the scale in embedded system with a smaller screen and resolution, can be achieved favorably using the Object-Oriented Discrete LOD Model. Meanwhile, this Model is much helpful to satisfy the needs of fast graphic display and highly efficient search. It is a load balancing solution on time and space and it is easy to maintain data uniformity. Thus it's suitable for embedded system which has limitations in hardware and it worth popularizing. In practical application, the Model can be perfected to the one that can dynamically define levels of details and critical scales to the user's appetite to give them a more appropriate response.
