A survey of road feature extraction methods from raster maps

Abstract

Maps contain abundant geospatial information, such as roads, settlements, and river networks, to name a few. The need to access this information to carry out analyses (e.g., in transportation, landscape planning, or ecology), as well as advances in software and hardware technologies, have driven the development of workflows to efficiently extract features from maps. The aim of this article is to provide a comprehensive overview of such methods to extract road features from raster maps. The methods are categorized based on the classes of techniques they employ (e.g., line extraction), as well as their subclasses (e.g., line tracing, Hough transform), the amount of user intervention required (e.g., interactive, automatic), the required data (e.g., scanned maps, contemporary vector data) and the produced results (e.g., raster‐based predictions, vector‐based results, attributes). Additionally, recent road extraction methods from overhead imagery, together with evaluation methods that will possibly benefit road extraction from raster maps, are reviewed. Furthermore, the evolution of this research field is analyzed over the past 35 years and the limitations of the current techniques, as well as possible future directions, are discussed.

A fast and effective deep learning approach for road extraction from historical maps by automatically generating training data with symbol reconstruction

“…Nowadays, this historical heritage has been digitalised into raster format through scanning and made widely accessible (Tsorlini et al, 2014). However, historical maps are subject to poor quality, which results from inaccurate surveying and reproduction technologies or chemical and physical deterioration (e.g., bleaching, paper distortion) (Jiao et al, 2021). Moreover, the scanning process could induce blurring and colour aliasing (Leyk et al, 2005;Liu et al, 2019;Uhl and Duan, 2021).…”

“…However, little attention has been paid to extracting roads from historical maps using deep learning. Compared to roads on remote sensing images presented in their natural form, roads on historical maps are represented by abstract symbols with various shapes and colours (Jiao et al, 2021). Although a tiny number of recent studies attempt to extract road features from historical maps using deep learning, such as road intersection (Saeedimoghaddam and Stepinski, 2020) and railroad (Chiang et al 2020), there is no research to systematically extract road networks from historical maps using deep learning.…”

Towards the automated large-scale reconstruction of past road networks from historical maps

“…More recently, scholars have applied deep learning methods such as convolutional neural networks (CNNs) to extract geometric and semantic transportation network characteristics from historical maps. These approaches include linear road feature extraction based on a U-Net CNN ( Ekim, Sertel, & Kabadayı, 2021 ; Jiao, Heitzler, & Hurni, 2021 ), extraction of road network intersections using an Inception-ResNet CNN ( Saeedimoghaddam & Stepinski, 2020 ), or road type recognition from cartographic road symbols using a U-Net CNN ( Can, Gerrits, & Kabadayi, 2021 ). Similarly, researchers have proposed deep learning based methods for the extraction of railroad networks ( Chiang, Duan, Leyk, Uhl, & Knoblock, 2020a ; Hosseini, McDonough, van Strien, Vane, & Wilson, 2021 ; Hosseini, Wilson, Beelen, & McDonough, 2021 ) from historical maps.…”

“…These deep-learning based methods are resource-intensive and require large amounts of typically manually labelled training data or templates. Jiao, Heitzler, and Hurni (2021) provide a detailed overview of these methods.…”

A Novel Data Augmentation Method to Enhance the Training Dataset for Road Extraction From Swiss Historical Maps

“…Chiang et al (2020) report a set of experiments for railroad extraction from USGS historical maps to investigate the impact of deep CNN architectures on feature extraction accuracy. Despite of the rapid development and the superiority in image segmentation and feature recognition of deep CNNs, their application to road extraction from historical maps is to some extent limited up to now (Jiao et al, 2021).…”