A Template Matching Based Approach for Geolocating Cadastral Aerial Images

Article Type

Research Article

Publication Title

SN Computer Science

Abstract

The rapid increase in aerial image analysis has created a compelling need to accurately determine the geolocation of regions of small, cropped images from a large context aerial image. The challenge lies in matching any land area captured from a drone or satellite to their corresponding locations within larger, full-scale cadastral maps. This is a task complicated by variations in scale, resolution, and land cover characteristics. To address this problem, an aerial image segmentation dataset and a cadastral map dataset were introduced by first segmenting cropped aerial images into distinct land cover classes such as Buildings, Roads, Vegetation, Trees, and Barren areas. These segmentations are then transformed into simplified cadastral maps highlighting critical boundaries using a deep learning-based multi-class segmentation method. Consecutively, we process the full-scale cadastral map along with randomly selected patches through a novel Deep Feature Template Matching Network (DFTMN). This network is designed to learn the correspondence between local image patches and their global context within the larger map, effectively aligning fine-grained features with their broader spatial surroundings. The proposed framework demonstrates strong potential for applications in urban planning, environmental monitoring, and disaster response, offering an effective solution for fine-to-coarse scale geolocation in aerial imagery. The code is available on https://github.com/asadidraco/GeolocationGitHub.

DOI

10.1007/s42979-025-04404-4

Publication Date

10-1-2025

Link to Full Text

Share

COinS