An automated unmanned aerial vehicle – structure from motion pipeline to create analysis-ready digital surface models for coastal monitoring
(2021) In Student thesis series INES NGEM01 20211Dept of Physical Geography and Ecosystem Science
- Abstract
- Anthropogenic pressure on coastal areas continues to increase, thus intensifying coastal erosion. Coastal monitoring programs are therefore essential to prevent severe ecosystem and economic consequences. Successful coastal monitoring schemes require highly accurate analysis-ready Digital surface models (DSMs) derived using the same workflow.
Using a Unmanned Aerial Vehicle – Structure from motion (UAV-SfM) approach allows for a cost-efficient, automated approach ensuring repeatability. Image segmentation is used to create water masks automatically, and an algorithmic approach to identify and import GCPs was developed. While most literature features a semi-automatic approach, this pipeline presents a workflow that allows for a fully... (More) - Anthropogenic pressure on coastal areas continues to increase, thus intensifying coastal erosion. Coastal monitoring programs are therefore essential to prevent severe ecosystem and economic consequences. Successful coastal monitoring schemes require highly accurate analysis-ready Digital surface models (DSMs) derived using the same workflow.
Using a Unmanned Aerial Vehicle – Structure from motion (UAV-SfM) approach allows for a cost-efficient, automated approach ensuring repeatability. Image segmentation is used to create water masks automatically, and an algorithmic approach to identify and import GCPs was developed. While most literature features a semi-automatic approach, this pipeline presents a workflow that allows for a fully automated DSM generation from UAV images. An image segmentation model (VGG-Segnet) is trained to automatically identify water and land areas in the UAV images resulting in a pixel accuracy of 90%. Ground control points (GCPs) are automatically identified using only the RGB images of the UAV by differentiating pixel clusters by color, size, and shape, as well as relative position.
While the outcome was not flawless and the markers were not always placed perfectly in the center, the approach showed high potential to be developed further. The study further shows the importance of using appropriate settings in Agisoft Metashape. Settings are likely to depend on equipment and study area. However, some insight into the effect of settings on the alignment quality is presented. The DSMs created in this study achieved an RMSE of 3 - 4 cm, proving a very high accuracy. Further analysis showed that this error is likely to be underestimated due to the poor distribution of check points. (Less) - Popular Abstract
- As human population continues to increase near coasts, coastal erosion events have become more severe. It is therefore important to monitor the coastline to prevent human loss and minimize economic and environmental consequences. To do this, a model of the study area needs to be created detailing the elevation as accurately as possible, a digital surface model. Individual digital surface models from different times can then be compared to identify the effect coastal erosion has upon a study area if an identical workflow is used.
Drones provide a both flexible and cost-efficient method to capture a study area. These individual images can then be ‘stitched together’ using photogrammetry to derive the elevation of the study area. By fully... (More) - As human population continues to increase near coasts, coastal erosion events have become more severe. It is therefore important to monitor the coastline to prevent human loss and minimize economic and environmental consequences. To do this, a model of the study area needs to be created detailing the elevation as accurately as possible, a digital surface model. Individual digital surface models from different times can then be compared to identify the effect coastal erosion has upon a study area if an identical workflow is used.
Drones provide a both flexible and cost-efficient method to capture a study area. These individual images can then be ‘stitched together’ using photogrammetry to derive the elevation of the study area. By fully automating this workflow, repeatability can be ensured making the digital surface models comparable. Besides writing a python script, it was required to mask out water and identify ground control points to fully automate the workflow. Ground control points describe points on the ground with known coordinates that are visible on the images. This allows to transform image, pixel, coordinates to real-world coordinates. Water moves constantly, making it difficult to calculate elevation accurately using photogrammetry. Artificial Intelligence was therefore used to train a model capable of identifying water in a drone image. Lastly, a multitude of settings within the photogrammetry software were analysed to optimize the accuracy of the digital surface models created.
The final digital surface models achieved a root mean square error of 3 – 4 cm, proving a very high accuracy and confirming high accuracies achieved in other studies using similar approaches. It was further shown that it is very important to use appropriate settings within the photogrammetry software Agisoft Metashape. When using the fully automated approach described in this paper, the automatic identification of ground control points remains a difficulty. The results showed that the approach is promising, but the points could not be identified as accurate as when identifying them manually. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9066921
- author
- Huber, Maximilian Rafael LU
- supervisor
- organization
- course
- NGEM01 20211
- year
- 2021
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- physical geography, ecosystem analysis, geomatics, GIS, UAV, structure-from-motion, machine learning, coastal monitoring
- publication/series
- Student thesis series INES
- report number
- 561
- language
- English
- id
- 9066921
- date added to LUP
- 2021-10-18 16:18:22
- date last changed
- 2022-01-01 03:40:05
@misc{9066921,
abstract = {{Anthropogenic pressure on coastal areas continues to increase, thus intensifying coastal erosion. Coastal monitoring programs are therefore essential to prevent severe ecosystem and economic consequences. Successful coastal monitoring schemes require highly accurate analysis-ready Digital surface models (DSMs) derived using the same workflow.
Using a Unmanned Aerial Vehicle – Structure from motion (UAV-SfM) approach allows for a cost-efficient, automated approach ensuring repeatability. Image segmentation is used to create water masks automatically, and an algorithmic approach to identify and import GCPs was developed. While most literature features a semi-automatic approach, this pipeline presents a workflow that allows for a fully automated DSM generation from UAV images. An image segmentation model (VGG-Segnet) is trained to automatically identify water and land areas in the UAV images resulting in a pixel accuracy of 90%. Ground control points (GCPs) are automatically identified using only the RGB images of the UAV by differentiating pixel clusters by color, size, and shape, as well as relative position.
While the outcome was not flawless and the markers were not always placed perfectly in the center, the approach showed high potential to be developed further. The study further shows the importance of using appropriate settings in Agisoft Metashape. Settings are likely to depend on equipment and study area. However, some insight into the effect of settings on the alignment quality is presented. The DSMs created in this study achieved an RMSE of 3 - 4 cm, proving a very high accuracy. Further analysis showed that this error is likely to be underestimated due to the poor distribution of check points.}},
author = {{Huber, Maximilian Rafael}},
language = {{eng}},
note = {{Student Paper}},
series = {{Student thesis series INES}},
title = {{An automated unmanned aerial vehicle – structure from motion pipeline to create analysis-ready digital surface models for coastal monitoring}},
year = {{2021}},
}