Comparative Study of Positional Accuracy Evaluation of Line Data
Yoshiaki Kagawa, Yoshihide Sekimoto and Ryosuke Shibaski
Center for Spatial Information Science and
Institute of Industrial Science,
University of Tokyo
4-6-1 Komaba, Meguro-ku, Tokyo 153-8505, Japan
Tel: +81-3-5452-6417 Fax: +81-3-5452-6417
Email:-kagawa@skl.iis.u-tokyo.ac.yp
Key word Line data, positional error, evaluation method, comparative analysis
Abstract
Recently, geographic information system (GIS) is used in a variety of fields. Spatial data is not usually free from errors and uncertainties, and they may have serious impacts on results of spatial query and analysis. In order to understand and evaluate these influences of errors and uncertainties, the quality of the spatial data needs to be evaluated quantitatively. In case of point data, it is easy to define the positional deviation or errors. But with regard to line data such as road boundary, not many measures or indicators have been proposed to characterize the positional error. In this study, the authors propose several measures to evaluate quantitatively the positional errors of line data, and compare the measure including the existing ones using actual data.
1.Introduction
Recently, geographic information system (GIS) is used intensively in a variety fields. And it is possible to acquire geographic data from various data sources. But, the data from different data sources are not always consistent due to errors. There are many sources of errors; e.g. errors from difference of primary acquisition methods, errors of digitizing, and errors from the lack of revision. It is important that we know what errors the data contain, and for that purpose, it is necessary to provide a set of methods for quantitative evaluation and to allow users to select evaluation methods according to their needs.
In the case of point data, we can easily evaluate the error by measuring root mean square error between corresponding point data. But in the case of line data, it is hard to identify the corresponding points. In this study, we propose evaluation methods for positional errors of line data, apply them to actual line data, clarify their characteristics.
2. Evaluation methods
Line data is represented by a series of point data. In the case of point data, we can identify which point data we measured. In the case of line data, however, we cannot identify which specific points along the line are measured. That is, it is difficult to identify the correspondence between "true" point and measurement point. The following sections describe proposed and existing methods for evaluation.
2.1 Buffer method
Buffer method was proposed by Shibasaki (shibasaki, 1992) based on Peucker's
e model. (Peucker, 1976) And Goodchild shows that distribution of errors make Gaussian distribution with this method (Goodchild, 1997). Consider a buffer of width
e around the true lie, then we search
e that contain a certain proportain (e.g. 95%) of measurement line's length within the buffer (Fig 1). This method investigates in how degree of range from true line measurement points exist, and does'nt, and identify errors explicity on the point-to-point basis. Consequently, it is difficult to compute likelihood of a line passing certain place. And bias of errors is difficult to estimate with this method. At the same time this method is insensitive to large errors from blunder.
Figure 1: Evaluation using buffer method
