Large amounts of map sheets and drawings are archived in society as scanned raster images or paper maps (that could be scanned). Examples include zoning maps or property drawings. These are often used as reference materials in urban planning or renovations. However, in order to use them as reference materials, they need to be digitized, which can be a time-consuming manual task. This project aims to automate the process that would facilitate the digitization and make old map data more easily accessible and reusable.
More information (in Swedish): https://www.oru.se/samverkan/oru-innovation/ai-impact-lab/avslutade-projekt/urbanaize/
Web demo as a result of this project: https://urbanaize.aass.oru.se/
In 2020, Statistics Sweden (SCB) scanned approx. 8,000 map sheets from the Population and Housing Censuses (FoB) in 1975 and 1980, respectively. Most of the map sheets are black-and-white economic sheets with delineation lines drawn in color for urban areas that were applied at the respective time. The map sheets are in the raster format TIF and sorted into folders based on county and municipality affiliation. An example of such a map image can be seen in Figure 1. The resolution for each map sheet is 300 dpi with approx. 7500 x 7000 pixels. The demarcations for the urban areas can be linked to statistics and contribute to a geographical description both of the rapid urbanization in connection with the million program and the breaking point of the so-called green wave. The material is, therefore, deemed to contain valuable information and relevant to digitize and make available as open data.
Figure 1. An example of a scanned map sheet as a result of the Population and Housing Censuses (FoB) in 1975. .
In this project, it is mainly the borders in FoB 1975 that are of interest. These borders have previously not been digitized and made available as vectorized geographical coordinates. Consequently, it is not possible to fully follow and understand Swedish urbanization and the geographical development of urbanization, which the scanned map material will be able to contribute to. The project aims to make available map material that otherwise risks being left unused in archives due to time-consuming manual digitization processes. Many actors in society would benefit from an automated digitization process as archived maps and drawings would become more accessible to community planners, construction companies, architects, individuals, and researchers alike, which could all benefit from accessible historical geographic data.
This project assumes the following project structure:
UrbanAIZE/
├── images/ # Images used for visualization
├── scripts/
├── __init__.py
├── annotation.py # Annotation of map images
├── dataset.py # Handles image pairs during training
└── map/
├── area.py # Support class and functions handle map areas
├── image.py # Helper class used to handle map image
└── utils.py # Util functions
└── model/
├── block.py # Common network block layers
├── unet.py # Implementation of the classic U-Net model
└── unet2plus.py # Implementation of the U-Net++ model
├── preparation.py # Preperation of map images
├── train.py # Traning of machine learning models
└── window/
├── annotation.py # Displays and handles annotation of map images
├── preparation.py # Displays and handles preperation of map images
└── window.py # Abstract joint window class
├── requirements.txt # Required Python libraries
├── .gitginore
├── LICENSE # MIT License
└── README
-
Clone this GitHub repository:
git clone https://github.com/statisticssweden/UrbanAIZE.git -
Install required Python libraries:
cd UrbanAIZE && pip3 install -r requirements.txt
As an initial procedure, download map images (in .tiff format), including the database file Tatorter_1980_2020.gpkg, and place them in a local folder. This folder's default name and location is ./data, but the folder's path and name can also be specified by the optional --path argument for all the operations listed below.
python3 scripts/preperation.py [--path <path>]
python3 scripts/annotation.py [--path <path>]
python3 scripts/train.py [--path <path>]
python3 scripts/prediction.py [--path <path>]