An Experiment on Distance Metrics Used for Road Matching in Data Integration
* Author to whom correspondence should be addressed.
Sigma Journal of Engineering and Natural Sciences 2016, Vol. 34, Issue 4, pp. 527-542; doi.org/10.62051/ytu.sigma-journal-of-engineering-and-natural-sciences-an-experiment-on-distance-metrics-used-for-road-matching-in-data-integration
Abstract
Keywords: Distance metrics; matching; integration; conflation; road network.
1. Introduction
Matching is a widely used procedure, especially in map conflation, updating and combining processes. It generally finds the similarities between objects in different datasets and creates relations between each object in order to analyze, combine, update, and transfer data. Also, it ensures the facilities about quality assesment for generalization (e.g. evaluating quality based on the distance between original and generalized lines) and analysis for travel behavior and route modelling (e.g. making positionally inaccurate real time GPS data significant and usable by matching with available road network data). In general terms, the matching procedure can be defined as the establishment of relations between different datasets that represent the same entities. The relations can be viewed as bridges that identify and connect the objects in different datasets, thereby enabling the datasets to be interoperable. Many algorithms have been developed to match spatial data from different sources automatically. Matching algorithms are usually designed to consider the similarities between objects. Geometry, topology or attribute similarities can be the components of the same similarity equations in complex matching processes [1,2]. Moreover, the features in a dataset can be matched to the features in another dataset using one, two, or all three criteria (i.e., geometries, topologies, and attributes) according to the complexity of the datasets. Any one of these criteria can be determined in the similarity equations. Feature type is highly important in selecting a matching method. The Euclidian distance in Equation(1), which is the shortest distance between points, can be used as a threshold for point-to-point matching; meanwhile, the Hausdorff distance in Equation(2), which is the longest of the shortest distances between lines, is suitable for line-toline matching [2,3]. The Hausdorff distance provides the spatial relationship between line features quantitatively for automated cartographic analyses. In addition to being used frequently in matching procedures for conflation processes, it is also employed to determine the amount of deviation of generalized lines from their original locations: 𝐷𝐸 = √(𝑋2 − 𝑋1 )2 + (𝑌2 − 𝑌1 )2 ,
The patterns and scales of the features should be considered in terms of geometric similarity. For example, a simple method that is adequate to match grid-patterned road networks may be inadequate to match circular or complex-patterned ones. For a building represented as a point in both datasets, Euclidian distance may be adequate for matching, while it may not be appropriate for the same building represented as an area in a larger scale dataset. Lynch and Saalfeld [4] defined map conflation as “Combining two digital map files to produce a third map file which is better than each of source maps.” In their initial attempts to solve the matching problem, the geometric incompatibilities between datasets were considered [4, 5, and 6]. Saalfeld [7] determined that the matching procedures are more successful with geometrically aligned data. Deretsky and Rodny [8] developed a method using the standard operators of relational database management systems on road lines, intersection points, and their attributes to integrate geometric and attribute data of two digital maps. Cobb et al. [9] developed a new method to match attribute-rich datasets in Vector Product Format developed by the National Imaging and Mapping Agency in the United States. In addition to conventional proximity, geometry, and topology, this method uses several semantics, such as data quality and feature code. This method is based on the semantic similarity of attributes and the shape similarity of lines, and it processes the data iteratively with respect to the selected criteria, starting from the strongest criterion and moving to the weaker ones until no new feature pairs are matched, like in Saalfeld’s [7] method. Walter and Fritsch [10] matched the road networks at close scales using statistical analysis. Their method initially aligns the road networks through affine transformation. Buffers are created around all of the road lines of the first dataset, and the road lines of the second dataset in these
An Experiment on Distance Metrics used for Road … / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
buffers are identified as matching candidates. The candidates are compared according to their locations, shapes, directions, and topologies. The roads to be matched are determined via the statistical analysis. Yuan and Tao [3] divided conflation into two general classes, as follows: “horizontal” (edgematching) and “vertical” (overlapping). These researchers highlighted the matching procedure as the hardest step. Kang [11] classified the matching procedure as “spatial” or “non-spatial.” He stated that spatial matching can use geometric measures and topological features for point, line, and area objects, while non-spatial matching can use statistical or rule-based methods. In their matching study on two datasets composed of polygons, Yuan and Tao [3] identified matches in the second dataset using the centroids in the first dataset through the raster shrink method. In this way, they performed attribute transfer between matching areas. The method developed by Xiong and Sperling [12] matches road features semi-automatically. Point, segment, and edge matchings are performed automatically, while matching pairs are manually checked. This control step prevents mistakes from being repeated and improves performance and reliability. Samal et al. [13] compared the attributes of polygon features and determined similarities. However, they found that features with significantly different attributes could not be matched. To measure the matching ability of features in terms of shape, the buffers were analyzed, and the important landmarks for polygon features were identified to calculate similarities between datasets. Finally, a similarity score was determined for each feature. Zhang and Meng [14] matched the road layers in the Basis digital land cover model (DLM) and TeleAtlas using topology and the unsymmetrical buffer growing method. Following this, they integrated postal data, including building numbers, into TeleAtlas. Mustière and Devogele [1] studied road networks with different scales. According to their proposed method, the matching candidate nodes are initially determined in a simple fashion according to the data-dependent distance threshold, and the matching candidate arcs are determined according to the Hausdorff distance. In this stage, a semi-Hausdorff distance between two arcs that is calculated from the most detailed to the least detailed one is used instead of the conventional Hausdorff distance. To determine the matched node pairs, specific measures are used, such as the network topology and the clockwise ordering of candidate arcs between the candidate nodes. The closest path approach is employed to determine the matched arcs. In this study, the success of the matching procedure could not be determined satisfactorily because both datasets came from same source (i.e., the French National Institute of Geographic and Forest Information) even though their scales were different. This method was enhanced in further studies [15,16] using belief theory and applied to spatial data from different sources. Kim et al. [17] matched the polygon features in large-scale datasets that included buildings using Voronoi diagrams, triangulation, and geometric measurements. Identifying the landmarks from attribute data, Voronoi cells were established, and then triangulation was performed. The geometric measures of triangles (i.e., area and circumference) were compared to calculate similarities and conduct the matching procedure. Li [18] and Li and Goodchild [19] developed an optimization model for the matching procedure. This model calculates total similarities for the features to be matched and matches the relevant features based on affine transformation parameters that maximize the similarities. Song et al. [20] carried out the matching procedure by creating a confidence matrix. A distance threshold was determined based on the maximum shift between two datasets. Points in the second dataset that were closer to each other than the set distance threshold were considered matching candidates. The number of lines connecting to the point in the first dataset was compared to the number of lines connecting to each matching candidate point. In the confidence matrix, a value of “1” was assigned to each matching candidate point with an equal number of
M. Hacar, T. Gökgöz / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
lines in the first dataset. After two iterations with the determined compatibility function, the points with maximum confidence values were matched. Pourabdollah et al. [21] performed a matching procedure to increase the quality of open data. The Open Street Map data, which are constantly updated but are deficient in terms of attributes, were conflated with the attribute-rich Ordnance Survey data to obtain an up-to-date and attributerich dataset. Yang et al. [22] classified the lines constituting the road network according to their patterns (i.e., atomic patterns and composite patterns) to determine matching candidates. In this method, the lines meeting the conventional geometrical and topological criteria are matched when they are in the same pattern class. Bierlaire et. al. [23] proposed a probabilistic map matching approach to relate GPS data with road network to generate meaningful paths. In the approach, after a set of potential true paths are generated, a likelihood with each of them is determined. In order to calculate the likelihood of the data for a specific path, both GPS coordinates and temporal information (i.e. speed and time) are used. Fan et. al. [24] proposed to match the OpenStreetMap road network with authority data. In the algorithm, first urban blocks, represented by polygons surrounded by their surrounding streets, are extracted. Then, by checking the topologies, the algorithm assigns road lines to edges of urban blocks. In the matching process, overlapping areas of blocks are used to match these polygons during the first step. In the second step, after edges of a matched polygon pair are matched with each other, road lines assigned to the same matched pair of urban block edges are matched with each other. Kang et. al. [25] developed an algorithm to associate geometric relationships between sidewalk and street segments. The algorithm contains three parameters: the distance between streets and sidewalk segments, the angle between sidewalk and street segments, and the difference between the lengths of matched sidewalk and street segments. In the literature there have been many studies about matching algorithms. Lots of them present ready-to-use algorithms with multiparameters. However, in this paper, matching process was carried out with only distance metrics to evaluate distance parameters. The present aimed to examine the five distance metrics in road matching (i.e., Euclidean distance between points, Euclidean distance between centroids, minimum Euclidean distance from point to line, Hausdorff distance, and semi-Hausdorff distance). An experiment was conducted using a plugin from an open desktop software program using two road datasets in different databases. The results were evaluated according to manual matching.
2. Matching Methods Using Distance Metrics
In this section five distance metrics were explained with mathematical equations and geometrical representations. In order to carry out the matching processes with these distances, user needs to specify a threshold (T) to determine certain matching pairs. Once T is specified by the user, the steps below are carried out for each method.
2.1. Matching Using the Distance between Points
Step 1: Lines composed of the same number of segments in different datasets are marked as the matching candidates. For example, the lines m and n in Figure 1 are composed of two segments, while line p comprises a single segment. Therefore, the lines m and n are marked as the matching candidates.
An Experiment on Distance Metrics used for Road … / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
Figure 1. Matching candidates (lines n and m). Step 2: Euclidian distances 𝑑𝑚𝑘−𝑛𝑙 are calculated with Equation(1) between all the points of line m and all the points of line n (Figure 2). Here, k and l are the numbers of points for lines m and n, respectively.
Figure 2. Euclidean distances (thin gray lines) between all points of lines m and n. Step 3: The shortest distance at each point is determined (𝑑𝑚1−𝑛1 , 𝑑𝑚2−𝑛2 and 𝑑𝑚3−𝑛3 ; Figure 3).
Figure 3. Shortest Euclidean distance (thin gray line) at each point. Step 4: The maximum of the shortest Euclidian distances is determined (𝑑𝑚𝑎𝑥 = 𝑑𝑚1−𝑛1 ) (Figure 4).
Figure 4. Maximum of the shortest Euclidian distances (𝑑𝑚𝑎𝑥 ; thin gray line). 531
M. Hacar, T. Gökgöz / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
2.2. Matching Using the Distance between Centroids
As the scale changes, geometries of the objects also change. A centroid is the geometrical parameter that is least affected from the scale changes. Thus, a centroid may be a good measure to match data with different scales. Step 1: Centroid coordinates of each line in the source and target datasets are calculated using the following formula: 𝑆𝑥
Here, 𝑋𝑔 and 𝑌𝑔 are the centroid coordinates of a line, 𝑆𝑖 is the length of segment i of the line, 𝑥𝑖 and 𝑦𝑖 are the midpoint coordinates of the segment i, w is the total number of segments, and S is the total length of the line. Figures 5a and b show the midpoints and centroids of the segments and lines, respectively.
Figure 5. (a) Midpoints of segments and (b) centroids of lines (black points). Step 2: The Euclidian distances between the centroids of two candidate lines are calculated. Step 3: The lines with a Euclidian distance that is smaller than the threshold T are matched.
2.3. Matching Using the Minimum Distance from a Point to a Line
Step 1: The shortest distances between the points of each line in a dataset and the lines in the other dataset are calculated. Here, 𝑑𝑚𝑘−𝑛 is the shortest distance between point k of line m and line n, while 𝑑𝑛𝑙−𝑚 is the shortest distance between point l of line n and line m. For example, as shown in Figure 6, the shortest distances between line m and line n are 𝐷𝑚−𝑛 = {𝑑𝑚1−𝑛 , 𝑑𝑚2−𝑛 , 𝑑𝑚3−𝑛 } and 𝐷𝑛−𝑚 = {𝑑𝑛1−𝑚 , 𝑑𝑛2−𝑚 , 𝑑𝑛3−𝑚 }. As there are no real perpendiculars from the points 1 and 3 of line n to line m, the lines connecting points 1 and 3 of line n to points 1 and 3 of line m, respectively, represent the shortest distances between these points. However, it is possible to draw perpendicular lines from points 1 and 3 of line m to line n. Thus, the shortest distances at points 1 and 3 of line m are the lengths of the perpendiculars from these points to line n. Similarly, while it is possible to draw a perpendicular from point 2 of line n to line m, it is not possible from point 2 of line m to line n. Thus, while the shortest distance at point 2 of line n is the perpendicular distance, the shortest distance at point 2 of line m is the length of the line connecting point 2 of line n to point 2 of line m.
An Experiment on Distance Metrics used for Road … / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
Figure 6. The shortest distances between points and lines (thin gray line). Step 2: The minimum of the shortest distances (𝑑𝑚𝑖𝑛 = 𝑚𝑖𝑛(𝐷𝑚−𝑛 , 𝐷𝑛−𝑚 ) is determined. For example, as shown in Figure 7, the shortest distance calculated at point 3 of line m (𝑑𝑚3−𝑛 ) is determined as the minimum of the shortest distances.
Figure 7. The minimum of the shortest distances (thin gray line). Step 3: If 𝑑𝑚𝑖𝑛 < 𝑇, the lines are matched.
2.4. Matching Using the Hausdorff Distance
Step 1: The shortest distances between the points of each line in a dataset and the lines in the other dataset are determined as in the “matching using the minimum distance from a point to a line” method. Step 2: The maximum of the shortest distances (𝑑𝑚𝑎𝑥 = 𝑚𝑎𝑥(𝐷𝑚−𝑛 , 𝐷𝑛−𝑚 )) is determined. For example, as shown in Figure 8, the maximum of the shortest distances between lines m and n is the distance between point 1 of line n and point 1 of line m (𝑑𝑛1−𝑚 ).
Figure 8. Maximum of the shortest distances (thin gray line). Step 3: If 𝑑𝑚𝑎𝑥 < 𝑇, the lines are matched.
2.5. Matching Using the Semi-Hausdorff Distance
This is the matching procedure based on the one-way Hausdorff distance from the line in the source dataset to the line in the target dataset.
M. Hacar, T. Gökgöz / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
Step 1: The shortest distances between the points of each line in the source dataset and the lines in the target dataset are determined. For example, as shown in Figures 6a and 6b, the shortest distances are determined from line n to m and from line m to n, respectively. Step 2: The maximum of the shortest distances (𝑚𝑎𝑥(𝐷𝑚−𝑛 ) or 𝑚𝑎𝑥(𝐷𝑛−𝑚 )) is determined. For example, while 𝑚𝑎𝑥 (𝐷𝑛−𝑚 ) = 𝑑𝑛1−𝑚 is the semi-Hausdorff distance from line n to line m (i.e., line m in the target and line n in the source datasets; Figure 8), 𝑚𝑎𝑥(𝐷𝑚−𝑛 ) = 𝑑𝑚2−𝑛 is the semi-Hausdorff distance from line m to line n (i.e., line n in the target and line m in the source datasets; Figure 9).
Figure 9. The semi-Hausdorff distance from line m to line n (thin gray line). Step 3: If 𝑚𝑎𝑥(𝐷𝑚−𝑛 ) < 𝑇 or 𝑚𝑎𝑥(𝐷𝑛−𝑚 ) < 𝑇, the lines are matched.
3. Experiment And Results
An experiment was conducted using the methods mentioned above with two road datasets in different databases, and the results were compared to manual matching results assumed to be the expected results to determine the numbers of correct and incorrect matchings. The methods were automatically implemented using MatchingPlugin (MatchingPlugIn0.7.2) from the OpenJump desktop software [26]. Two datasets composed of several road centerlines in a part of the Beykoz district (10 km x 6 km) of Istanbul, Turkey were used (Figure 10). The datasets were produced by the Istanbul Metropolitan Municipality (IMM) Directorate of Geographical Information Systems and Başarsoft Information Technologies Inc. (Figure 11). The study area was further divided into three zones to facilitate the verification of automatic matching (Figure 12). Hereafter, the IMM data are referred to as the “target dataset,” and the Başarsoft data are labelled the “source dataset.”
Figure 10. The study area (red rectangle) and inset map (Map data was taken from GADM database [27]). 534
An Experiment on Distance Metrics used for Road … / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
Figure 12. Both IMM (target, green) and Başarsoft roads (source, red) and zones. 535
M. Hacar, T. Gökgöz / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
While the target dataset is used in the Geographic Information System (GIS) infrastructure by IMM at scale of 1:1,000, the source dataset at scale of 1:5000 is used in navigation applications by Başarsoft. The datum of the target dataset is ITRF2005.0, and the datum of the source dataset is WGS84. There are locational and topological differences between the datasets due to the production method, date, source, scale, and so on (Figure 13). Some road names in the target dataset are not up-to-date. While the source dataset is attribute-rich, it is not geometrically up-todate (some roads do not exist), and some roads are represented with simpler geometry.
Figure 13. Overlapping view of the target (green) and source (red) datasets, and the samples showing the roads are not exactly superposed. While some roads are represented with two or more line features in the target dataset, they are represented with a single line feature in the source dataset (or vice versa). Thus, the number of matched features in manual matching was determined by considering 1:1, 1:N, and N:1 matching conditions (Table 1). To measure the success of the automated methods with respect to the manual matching, the numbers of “Matching,” “Correct Matching,” “Incorrect Matching,” “Incomplete Matching,” and “No Matching” features have been determined. Table 1. Numbers of manually matched features in each zone.
Number of features Number of matched features Number of unmatched features Number of matching pairs
In the experiment, 1:1 and 1:N matchings were tested for all methods by specifying the distance threshold as T=10 meters. The threshold value was determined by manually measuring the maximum distances between lines in the datasets. The results obtained in Zones 1, 2, and 3
An Experiment on Distance Metrics used for Road … / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
are presented as manual, automated, correct, incorrect, incomplete, and unmatched feature counts in Tables 2, 3, and 4, respectively. Table 2. Zone 1: 1:N and 1:1 matching results. Zone 1
Minimum distance from point to line 1:N 1:1 332 332 292 17 150 314 121 1 21 0 40
Minimum distance from point to line 1:N 1:1 248 248 208 22 126 226 67 0 15 0 40
Minimum distance from point to line 1:N 1:1 215 214 143 36 66 178 69 0 8 1 72
To determine the success of the matching methods, percentages of automated, correct, and incorrect matchings were calculated (Tables 5, 6, and 7). While 99.8% of the features were automatically matched with the method that used the minimum distance from a point to a line, which represented the highest matching percentage for 1:N matching (Table 5), it was found that most of the matchings were incorrect (Table 7). However, 58.0% of the features on average were correctly matched using the semi-Hausdorff distance method, which gave the best results for 1:1
M. Hacar, T. Gökgöz / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
matching (Table 6). Matching using the Hausdorff distance produced the least incorrect matchings. Moreover, the method using the distance between centroids produced almost the same results as the method using semi-Hausdorff distance for 1:1 matching. The method using the distance between centroids produced the next best results after the Hausdorff distance method in terms of incorrect matching (Table 7). Table 5. Percentages of automated matching ((𝐴𝑢𝑡𝑜𝑚𝑎𝑡𝑒𝑑⁄𝑀𝑎𝑛𝑢𝑎𝑙) × 100).
Distance between points Distance between centroids Minimum distance from point to line Hausdorff distance Semi-Hausdorff distance
Table 6. Percentages of correct matching ((𝐶𝑜𝑟𝑟𝑒𝑐𝑡⁄𝑀𝑎𝑛𝑢𝑎𝑙) × 100).
Distance between points Distance between centroids Minimum distance from point to line Hausdorff distance Semi-Hausdorff distance
Table 7. Percentages of incorrect matching ((𝐼𝑛𝑐𝑜𝑟𝑟𝑒𝑐𝑡⁄𝑀𝑎𝑛𝑢𝑎𝑙) × 100).
Distance between points Distance between centroids Minimum distance from point to line Hausdorff distance Semi-Hausdorff distance
3.1. Evaluation of the Results
In road-matching tasks, evaluation of the results depends on the purpose of matching. For example, vehicles like ambulances, firetrucks, mail trucks, school buses, and taxis use navigation systems that visualize road networks. These systems must always be up-to-date and reliable. In this regard, the accuracy of the matching method should reach a level that satisfies the requirements of the navigation system [28]. In this study, the methods were evaluated according to the correct and incorrect matchings.
An Experiment on Distance Metrics used for Road … / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
The results obtained in each zone with the methods using Hausdorff and semi-Hausdorff distances showed that these approaches cannot be used on their own. However, the fact that the number of incorrect matchings of these methods (especially using Hausdorff distance) was smaller than that of the others explains why these are used as one of the primary measures in most matching algorithms (Tables 2, 3, and 4). The method using distance between centroids is not a widely used method for line matching. However, it was the second best method in terms of incorrect matching, following the Hausdorff distance method. In addition, although this method was less successful than the method using semi-Hausdorff distance in terms of correct matchings, it was more successful than the method using Hausdorff distance (Tables 2, 3, and 4). In this context, while the distance between two centroids is not adequate as a measure for use in a matching method, it is a measure that can be employed in the same way as the Hausdorff distance in algorithms containing similarity equations with different criteria. The matching results obtained in Zone 1 are represented as a graph in Figure 14. It is obvious that the methods using distance between centroids and Hausdorff and semi-Hausdorff distances were more successful than the other approaches in terms of correct and incorrect matchings.
Figure 14. Zone 1: 1:1 matching results. As shown in Figure 15, all roads in Zone 1 that were incorrectly matched by all methods were positioned at the junctions. For example, a rectangular junction in the source dataset was represented as a triangular junction in the target dataset (Figure 16).
M. Hacar, T. Gökgöz / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
Figure 15. Positions of roads (red points) incorrectly matched by all methods in Zone 1.
Figure 16. Different representations of a junction in the source (red lines) and target (green lines) datasets overlapped with satellite image [29].
4. Conclusion And Discussion
Each method used in this study has its own characteristics. While an object pair is determined as matched in one of the five methods, it may be unmatched in another. Also, while a distance metric is relatively accurate for matching the datasets in one region, it may fail in other regions. This means that the distance metrics are data-dependent. Hausdorff and semi-Hausdorff distances are more satisfactory that are already used in many matching methods. In the literature, many of
An Experiment on Distance Metrics used for Road … / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
the researchers have prefered these distance metrics since the determination of the difference between two line is considered as from points of first line to the other line. However, the distance between the centroids of lines (from point to point) seems like a useful metric for road matching as well. It was the second best method in terms of incorrect matching, following the Hausdorff distance method. Also, none of the tested methods were completely satisfactory for road matching since the purpose of this study is only the examination of the distance metrics. The similarity equations determine the success of the matching processes directly. The study also shows all of the examined methods failed at junctions. This study was only implemented in datasets with similar large scales (1:1.000 and 1:5.000). To make a comparison between sources with multi-scale, it may be tested in datasets with middle scales (e.g. 1:25.000 and 1:100.000). Angle metrics may be examined to determine angle parameter. Different parameters such as angle, topology or attribute informations may be used together to create similarities between objects of datasets. Using the patterns of junctions in road matching may be a topic worth examining. Acknowledgments / Teşekkür Authors would like to thank Michaël Michaud (one of OpenJump project leaders) for his support, and IMM Directorate of Geographical Information Systems and Başarsoft Information Technologies Inc. for supplying road datasets. Authors are also grateful to the anonymous reviewers whose comments helped to improve the paper. REFERENCES / KAYNAKLAR [1] [2] [3] [4] [5] [6] [7] [8] [9]
Mustière, S. and Devogele, T., (2008). “Matching Networks with Different Levels of Detail”, GeoInformatica, 12(4): 435-453. Zhang, M., (2009). Methods and Implementations of Road-Network Matching. Ph. D. Thesis, Technical University of Munich, Münih. Yuan, S. and Tao, C., (1999). “Development of Conflation Components”, The Proceedings of Geoinformatics'99 Conference, 19-21 June 1999, Ann Arbor, 1-13. Lynch, M.P. and Saalfeld, A., (1985). “Conflation: Automated Map Compilation—A Video Game Approach”, Auto-Carto VII, 11-14 March 1985, Washington, D.C. Rosen, B. and Saalfeld, A., (1985). “Match Criteria for Automatic Alignment” AutoCarto VII, 11-14 March 1985, Washington, D.C. Lupien, A. E. and Moreland, W.H., (1987). “A General Approach to Map Conflation”, Auto-Carto VIII, 29 Mart- 3 April 1987, Baltimore. Saalfeld, A., (1988). “Conflation: Automated Map Compilation”, International Journal of Geographical Information System, 2(3): 217-228. Deretsky, Z. and Rdony, U., (1993). “Automatic Conflation of Digital Maps”, In Vehicle Navigation and Information Systems Conference, October 1993, Ottawa. Cobb, M.A., Chung, M.J., Foley III, H., Petry, F.E., Shaw, K.B. and Miller, H.V., (1998). “A Rule-Based Approach for the Conflation of Attributed Vector Data”, GeoInformatica, 2(1): 7-35. Walter, V. and Fritsch, D., (1999). “Matching Spatial Data Sets: A Statistical Approach”, International Journal of Geographical Information Science, 13(5): 445-473. Kang, H., (2002). Analytical Conflation of Spatial Data from Municipal and Federal Government Agencies, Ph. D. Thesis, the Ohio State University, Ohio. Xiong, D. and Sperling, J., (2004). “Semiautomated Matching for Network Database Integration”, ISPRS Journal of Photogrammetry and Remote Sensing, 59(1): 35-46.
M. Hacar, T. Gökgöz / Sigma J Eng & Nat Sci 34 (4), 527-542, 2016
Samal, A., Seth, S. and Cueto, K., (2004). “A Feature-Based Approach to Conflation of Geospatial Sources”, International Journal of Geographical Information Science, 18(5): 459-489. Zhang, M. and Meng, L., (2007). “An Iterative Road-Matching Approach for The Integration of Postal Data”, Computers, Environment and Urban Systems, 31(5): 597615. Olteanu-Raimond, A.M. and Mustière, S., (2008).“Data Matching—A Matter of Belief”, Proceedings of the International Symposium on Spatial Data Handling, 23–25 June 2008, Montpellier, 501-19. Olteanu-Raimond, A.M., Mustière, S. and Ruas, A., (2015). “Knowledge Formalisation for Vector Data Matching Using Belief Theory”, Journal of Spatial Information Science, 10: 21-46. Kim, J.O., Yu, K., Heo, J. and Lee, W.H., (2010). “A New Method for Matching Objects in Two Different Geospatial Datasets Based on the Geographic Context”, Computers & Geosciences, 36(9): 1115-1122. Li, L., (2010). Design of A Conceptual Framework and Approaches for Geo-Object Data Conflation, Ph. D. Thesis, University of California, Santa Barbara. Li, L. and Goodchild, M.F., (2011). “An Optimisation Model for Linear Feature Matching in Geographical Data Conflation”, International Journal of Image and Data Fusion, 2(4): 309-328. Song, W., Keller, J.M., Haithcoat, T.L. and Davis, C.H., (2011). “Relaxation‐Based Point Feature Matching for Vector Map Conflation”, Transactions in GIS, 15(1): 43-60. Pourabdollah, A., Morley, J., Feldman, S. and Jackson, M., (2013). “Towards An Authoritative Openstreetmap: Conflating OSM and OS Opendata National Maps’ Road Network”, ISPRS International Journal of Geo-Information, 2(3): 704-728. Yang, W., Lee, D. and Ahmed, N., (2014). “Pattern Based Feature Matching for Geospatial Data Conflation”, GEOProcessing, March 2014, Barcelona. Bierlaire, M., Chen, J., and Newman, J., (2013). “A probabilistic map matching method for smartphone GPS data”, Transportation Research Part C: Emerging Technologies, 26: 78-98. Fan, H., Yang, B., Zipf, A., and Rousell, A., (2016). “A polygon-based approach for matching OpenStreetMap road networks with regional transit authority data”, International Journal of Geographical Information Science, 30(4): 748-764. Kang, B., Scully, J. Y., Stewart, O., Hurvitz, P. M., and Moudon, A. V., (2015). “SplitMatch-Aggregate (SMA) algorithm: integrating sidewalk data with transportation network data in GIS”, International Journal of Geographical Information Science, 29(3): 440-453. Michaud M., MatchingPlugIn Tutorial for Version 0.7.2., [Internet] http://sourceforge.net/projects/jumppilot/files/OpenJUMP_plugins/More%20Plugins/Matching%20PlugIn/MatchingPlugIn0. 7.2.pdf/download, [Accessed on 15.03.2016]. Global Administrative Areas, GADM database, California, USA, [Internet] http://www.gadm.org/, [Accessed on 10.08.2016] Hacar, M., (2015). Mekânsal Veri Altyapilarinda Geometrik Entegrasyon, M. Sc. Thesis, Yıldız Technical University, İstanbul. Esri, DigitalGlobe, GeoEye, Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community.
Share and Cite
HACAR, M.; GÖKGÖZ, T. An Experiment on Distance Metrics Used for Road Matching in Data Integration. Sigma Journal of Engineering and Natural Sciences 2016, Vol. 34, pp. 527-542. https://doi.org/10.62051/ytu.sigma-journal-of-engineering-and-natural-sciences-an-experiment-on-distance-metrics-used-for-road-matching-in-data-integration