Quality assessment and object matching of OpenStreetMap in combination with the Dutch topographic map TOP10NL

More Info
expand_more

Abstract

The possibility of an automatic object matching process in combination with a VGI (Volunteered Geographical Information) type of map has been explored in this research. TOP10NL, the topographic map of the Netherlands and OpenStreetMap (OSM), the VGI type of map have been chosen for this purpose. The object matching process in this research became more critical as OSM does not always follow strict rules. Therefore, to overcome the complications special care has been taken into account on the characteristics of these two datasets and have been incorporated in the object matching process. Knowledge from previous studies in this field has been explored and adapted in this study. As TOP10NL and OSM represent completely different type of datamodels, a schema translation had to be done to harmonise them. The average geometric accuracy of TOP10NL is ±2 meter, whereas the quality of OSM is uncertain over locations. Therefore the quality of the OSM dataset over the study areas was also judged before developing the object matching model by some quality measures of geographic maps. Only the quality measures ‘data lineage’, ‘completeness’ and ‘positional accuracy’ were considered in this study. The data lineage shows that both datasets share quite a large number of objects from the same origin. It was found that the road, railway and building objects were quite complete in the OSM dataset over the study areas whereas the water objects were quite incomplete. Because of the reliable data lineage of the OSM data over the Netherlands the positional accuracy of the OSM data were found to be quite good. Considering the geographic location and the importance four areas within the Netherlands were chosen for the experiments. The urbanised Delft and Rotterdam, semi-urbanised Dokkum and the rural part of Echt were selected for the experiments. The common features within both TOP10NL and OSM, which are road, building, water and railway, were considered for the experiments. Two different object matching models were developed for line and polygon objects. For both models the decision rule was developed considering the characteristics of the datasets used in this study. For line matching the decision rule was developed to make the list of best matched pairs by accumulating them in a step by step procedure, whereas a straight forward simple rule has been developed for polygon matching process. These models were verified with the different object class data of the different study areas. The accuracy was high and for more than 90% instances the correct match was found. In the road matching process it was possible to insert the ‘street names’ from the OSM database as an attribute in the road network database of TOP10NL. A number of matching tables have been formed for different cardinality of matched objects which may be useful in the future in an automatic updating process of TOP10NL and OSM.