Conversion of Cadastral Survey Information into LandXML Files using Machine Learning

De La Rosa, Oscar Garrido (2019) Conversion of Cadastral Survey Information into LandXML Files using Machine Learning. [USQ Project]

[img]
Preview
Text (Project)
GarridodelaRosa_O_Campbell_Redacted.pdf

Download (1MB) | Preview

Abstract

Although new cadastral surveys can readily be produced in the industry standard LandXML format, there is a vast amount of pre-existing information which is only stored as image files. Automating the back-capture of this information would improve a process which is labour intensive and prone to human error. This project proposes a workflow to automate this process, in relation to Victorian cadastral survey information. Specific algorithms and outcomes are examined using a simplified sample cadastral plan.

The literature review reveals that similar documentation processes have been undertaken in other fields, such as music (Calvo-Zaragoza et al., 2018). In the cadastral context only true to scale cadastral maps have been digitised but not surveyors’ sketches or field records (Ignjatić et al., 2018)

A simple plan was created containing a closed parcel and two instrument points for creation and testing of the workflow. An analysis of the tasks required to extract the information needed for the LandXML files was undertaken. A pipeline was designed to perform the data extraction in a machine learning environment, which has been dubbed Double Filter Capture. It consists of two main workflows that handle the graphical information and the text elements separately, by means of Computer Vision and Optical Character Recognition algorithms, respectively. An implementation of the actions in the pipeline was trialled and barriers encountered discussed. Several Machine Learning algorithms were used for the required tasks, such as line detection, corner detection, image rotation, text detection and text extraction.

The project gives some idea of the possibilities and limitations that a larger scale automated back-capture would face, when dealing with records of significantly greater complexity. It also points the way to further research required to refine the extraction process outlined here, for example including elements omitted in this project, such as occupation and other auxiliary information and hand-written records.

This project demonstrates automated accurate data extraction from an image file is possible, however an extensive investment would be required in the programming stage, given the complexity and inconsistencies of existing plans that require back-capture.


Statistics for USQ ePrint 43149
Statistics for this ePrint Item
Item Type: USQ Project
Item Status: Live Archive
Additional Information: Bachelor of Spatial Science (Honours)(Surveying)
Faculty/School / Institute/Centre: Current - Faculty of Health, Engineering and Sciences - School of Civil Engineering and Surveying (1 Jul 2013 -)
Faculty/School / Institute/Centre: Current - Faculty of Health, Engineering and Sciences - School of Civil Engineering and Surveying (1 Jul 2013 -)
Supervisors: Campbell, Glenn
Date Deposited: 18 Aug 2021 06:33
Last Modified: 18 Aug 2021 06:33
Uncontrolled Keywords: Cadastral surveying, computer vision, machine learning, LandXML
URI: http://eprints.usq.edu.au/id/eprint/43149

Actions (login required)

View Item Archive Repository Staff Only