Cartographic reconstruction of building footprints from historical maps: A study on the Swiss Siegfried map

Abstract

Extracting features from printed maps has been a challenge for decades; historical maps pose an even larger problem due to manual, inconsistent drawing or scribing, low printing quality, and geometrical distortions. In this article, a new workflow is introduced, consisting of a segmentation step and a vectorization step to acquire high‐quality polygon representations of building footprints from the Siegfried map series. For segmentation, an ensemble of U‐Nets is trained, yielding pixel‐based predictions with an average intersection over union of 88.2% and an average precision of 98.55%. For vectorization, methods based on contour tracing and orientation‐based clustering are proposed to approximate idealized polygonal representations. The workflow has been tested on 10 randomly selected map sheets from the Siegfried map, showing that the time required to manually correct these polygons drops to about 45 min per map sheet. Of this sample, approximately 10% of buildings required manual corrections. This workflow can serve as a blueprint for similar vectorization efforts.

A Geomatic Approach to the Preservation and 3D Communication of Urban Cultural Heritage for the History of the City: The Journey of Napoleon in Venice

“…For the choice of the mathematic model of transformation, the characteristics, in terms of accuracy, of the historical cartography adopted and the extent of the area under analysis have been taken into account [44,45]. It was therefore decided to adopt a transformation model present in the software ArcGIS Pro called "adjust", since it integrates global and local transformation.…”

Cartographic Relief Shading with Neural Networks

“…Our final network is a U-Net [10]. U-Nets use connected encoding and decoding layers and have proven to be particularly useful for image segmentation, e.g., in the context of biomedical images [10] or geospatial raster data [64]. A U-Net consists of a contracting path (or encoder) to analyse the spatial context of an image and an expanding path (or decoder) that allows for accurate localisation of predicted features.…”

Combining Remote-Sensing-Derived Data and Historical Maps for Long-Term Back-Casting of Urban Extents

“…Such recent efforts include the mining of (historical) map collections by their content or associated metadata [ 32 - 37 ], automated georeferencing [ 18 , 38 - 40 ] and alignment [ 41 , 42 ], text detection and recognition [ 43 - 45 ], and the extraction of thematic map content, often involving (deep) machine learning methods, focusing on specific geographic features such as forest [ 46 ], railroads [ 33 , 47 ], road network intersections [ 48 , 49 ] and road types [ 50 ], archeological content [ 51 ] and mining features [ 52 ], cadastral parcels boundaries [ 53 , 54 ], wetlands and other hydrographic features [ 55 , 56 ], linear features in general [ 57 ], land cover/land use [ 58 ], urban street networks and city blocks [ 34 ], building footprints [ 13 , 59 , 60 ], and historical human settlement patterns [ 61 - 63 ]. Other approaches use deep-learning-based computer vision for generic segmentation of historical maps [ 64 , 65 ], generative machine learning approaches for map style transfer [ 66 , 67 ], or attempt to mimic historical overhead imagery based on historical maps [ 68 ].…”