UAV Image Acquisition Using Structure from Motion to Visualise a Coastal Dune System

INTRODUCTION

Coastal zones and the geomorphology of this environment have been of great interest to historians, scientists, and the public for many centuries, mainly due to the large number of people across the world who dwell near to the coast. The knowledge gained in understanding the impacts of a gain or reduction in beach and dune systems, either naturally or by human inducement, is vital as to the role that these fragile regions have in the future. The presence of dune systems must be seriously considered, as many are the first and last defence that rural environments, towns, and cities have against coastal inundation and flooding.

Traditional approaches to coastal monitoring have long been the reserve of airborne methods such as Light Detection and Ranging (LiDAR), Radio Detection and Ranging (RADAR), terrestrial laser scanning (Micheletti et al., 2014, 2015; Fabbri et al., 2017), historic and contemporary aerial photography (Alberico et al., 2017), and satellite imagery (Gomez et al., 2014; Papakonstantinou et al., 2016; McCarthy et al., 2017) to collect remotely sensed data. Many of these methods will generate high-quality point clouds and digital surface models (DSMs), nevertheless there may be significant cost implications, lengthy processing times, georeferencing, and accuracy of vertical elevations (Klemas, 2015). These remote sensing techniques have been important developments for mapping large areas of coastline that are remote, yet they all share drawbacks including lengthy processing times, expensive methods of data capture, and lack the spatial accuracy of a survey closer to the ground (Drummond et al., 2015). Professional standard Unmanned Aerial Vehicles (UAVs) have been around for many years, and applications for their deployment have been growing, mostly developed and utilised by the military, film and television industry, structural and building survey, parcel delivery, surveillance, and more recently for emergency disaster response (Shakhatreh et al., 2018). Successful UAV surveys have been completed on sections of river bank, agricultural applications, coastal salt marshes, estuary vegetation mapping, and coastal dune systems, at low altitudes (<100m) have been showrn to generate very high pixel resolution <5 cm/pixel (Mancini et al., 2013; Drummond et al., 2015; Marteau et al.. 2016; Papakonstantinou et al., 2016; Woodget et al., 2017), which allows for precise mapping compared to traditional remotely sensed imagery, which at very best may range from 10 to lOOmVpixel (Gomez et al., 2014).

The UAV approach for image capture and analysis within a dynamic coastal environment has only recently been shown as a viable option to aid with mapping and monitoring as the accuracy that can be achieved is significant, compared to traditional methods (Drummond et al., 2015; Moloney et al., 2018). UAV advancements in recent years have included miniaturisation of hardware and software, improved resolution of sensors, battery efficiencies, continued advancements with automated flight, and possibly the most key advancement is the reduction in price (Klemas, 2015; Marteau et al., 2016). Technological advancements within the control systems of UAVs can now be achieved by the existing and improved autonomous flight planning and new technology such as using virtual reality (VR) smart glasses as a control (replacing VR headsets and first-person view). Operation of termed ‘off-the-shelf UAVs (Klemas, 2015; Turner et al., 2016; Moloney et al., 2018) has revolutionised the capacity of individuals and consortiums to enter the survey industry, disaster response, and environmental protection, chiefly through the reduced cost of these UAVs. UAV environmental coastal survey has for many years been conducted by previously mentioned aerial survey techniques which are expensive to employ and are not always guaranteed to achieve clear imagery. Therefore, this type of image acquisition has been largely used by large commercial companies and governments for mapping and monitoring (Moloney et al., 2018).

Structure from Motion SfM (SFM) owes much of its existence to mathematical photogramme- try models developed in the 1950s and 1960s, which were able to demonstrate spatial relationships between images (Micheletti et al., 2015). The traditional SfM (SFM) photogrammetry approach has been around for many years and most notably used in architecture, urban landscapes, engineering, industrial survey, and within medical science and has been heavily influenced by the advancements in computing science (Schenk, 2005). SfM (SFM) applications used within architecture and

engineering are less reliant on absolute geographic location (ground control); therefore, unstructured overlapping images, obtained from multiple cameras, from a range of viewpoints, are favourable to generate three-dimensional (3D) visualisations and models (Westoby et al., 2012). SfM (SFM) algorithms within the software are designed to automatically image match common features between each image within the data, positioning and orientating parameters throughout the dataset. Following orientation, production of a high resolution and colour-coded (RGB) point cloud that represents the object is created. Micheletti et al. (2014) suggest that the larger the dataset (increase in the number of images), the easier it is to reject lower quality images, greatly improving the density of the point cloud and refining the accuracy of the model. SfM (SFM) techniques used as a tool for geographical or topographical survey have emerged from the advances in modelling software, computational power, and traditional photogrammetry (Warrick et al., 2017). SfM (SFM) software generates very accurate 3D DSMs and digital elevation models (DEMs) from two-dimensional (2D) imagery which are all powerful visualisation tools. The overriding objective of SfM (SFM) software is to reconstruct real- world 3D geometry utilising 2D photos that depict a static environment and recreating the environment within a 3D VR model. The development of SfM (SFM) within geosciences has provided many opportunities to generate accurate topographic maps and high-resolution and high-quality 3D results from low-cost data acquisition methods such as the UAV (Westoby et al., 2012; Nex and Remondino, 2014; Drummond et al., 2015; Turner et al., 2016; Moloney et al., 2018).

SfM (SFM) does not explicitly require the use of ground control points (GCPs), as many locations have easily identifiable reference points where georeferencing of the image can easily be obtained. The addition of GCP is essential within the coastal margin as fluctuations with the mean low and high water marks constantly change depending on the geographic location. Employing the GCP technique within the SfM (SFM) photogrammetry is essential to orientate the model (defining an absolute geographic location) and must be noted as coastal processes within these environments are subject to constant change from tidal dynamics and weather phenomena (Moloney et al.. 2018). Mancini et al. (2013) suggest that the Global Navigation Satellite System (GNSS) or Global Positioning Systems (GPSs) are valuable additions when mapping shoreline locations to accurately define direction (X and Y values), elevation (Z value), and for implementation of GCPs within the environment (Drummond et al., 2015). These spatial datasets can easily be added to a GIS package for analysis.

This project will utilise a multirotor UAV (Quadcopter) to conduct an autonomous overflight of the survey site, collect structured images covering a dune system, and process these within SfM (SFM) software (Pix4D Cloud®, 2018) to create point clouds, DSMs, orthomosaic imagery, and 3D models (Mancini et al., 2013; Nex and Remondino, 2014; Papakonstantinou et al., 2016; Turner et al., 2016). The importance of generating this up-to-date DSM and ground survey (GNSS) is to be able to test the reliability against alternative sourced data such as LiDAR and Ordnance Survey (OS) DSMs, as modelling for potential coastal flood inundations for many areas of coastline use these data types (Warrick et al., 2017). The structured approach to data collection will result in a large dataset (images) covering the whole study site, which hopes to significantly improve the resolution and accuracy of the visualisations (Micheletti et al., 2015). The pre-flight SfM (SFM) application approach to survey facilitates a pre-programmed autonomous flight path and height to be uploaded to the UAV which will result in a structured image capture process ensuring the required 65%-75% overlap of imagery (Pix4D, 2018).

The objectives within this project are to: [1] [2] [3]

4. Process aerial imagery using SfM (SFM) software

i. Comparison between Pix4D Cloud and Pix4Dmapper to generate DSM, orthomosaic, and 3D models

ii. Comparison between traditional methods (LiDAR) and SfM (SFM) DSM SURVEY LOCATION

The area of interest is within the Tentsmuir National Nature Reserve (under the stewardship of Scottish Natural Heritage (SNH)) situated at the north-east tip of Fife, which provides an excellent range of environments; mature dunes, mature slacks, mobile dunes, mobile and fixed ridges, and evidence of severe coastal erosion, all contained within an area of 35 hectares and just over a kilometre in length. This area of coastline is rich in soft sandy beaches and is part of an ancient sand dune system which dates back 6,000years and extends several kilometres to the west. Extensive geomor- phological processes to the dune system both positive and negative have been in evidence within this environment which has shaped the area that we see now (Cunningham, 2014). The reserve generates a large amount of tourism throughout the year including dog walkers, summer sun seekers, educational visits, and outdoor sports enthusiasts, which has undoubtedly led to damage to parts of the reserve. This is certainly a factor as to the speeded-up erosion of sections of path within the site. Three bands can be defined at this location: mature fixed dunes extending many kilometres to the west; dune slacks made up of mature dunes, hummocks, and moist fertile meadow or marsh; and mobile dunes on the seaward side (east) of the reserve that have migrated (west) over many tens of metres over the last 10 years.

  • [1] Identify and define survey area: i. Test effectiveness of mobile mapping application to identify points of interest (POIs)
  • [2] Define GCP locations within the survey area i. Conduct a GNSS survey of GCPs ii. Identify two smaller sites within the perimeter to conduct ground truths
  • [3] Conduct a full overflight of the area using an automated UAV platform
 
Source
< Prev   CONTENTS   Source   Next >