Video Object Detection for Maritime Navigation
(2024) In Master's Theses in Mathematical Sciences FMAM05 20241Mathematics (Faculty of Engineering)
- Abstract
- The purpose of this thesis was to investigate whether video object detection methods can outperform single-frame object detection methods in the context of maritime navigation. Object detection is a well-explored area within machine learning with many high performing published methods. Video object detection however, where information from previous video frames is considered in order to capture historical data such as movement in the detections, is a relatively recent area.
In this thesis I have researched current state-of-the-art within video object detection and applied these methods to the data provided by Saab Kockums. The video object detection methods can successfully detect navigation objects in video, even small and far away... (More) - The purpose of this thesis was to investigate whether video object detection methods can outperform single-frame object detection methods in the context of maritime navigation. Object detection is a well-explored area within machine learning with many high performing published methods. Video object detection however, where information from previous video frames is considered in order to capture historical data such as movement in the detections, is a relatively recent area.
In this thesis I have researched current state-of-the-art within video object detection and applied these methods to the data provided by Saab Kockums. The video object detection methods can successfully detect navigation objects in video, even small and far away objects that single-frame object detection methods have difficulties finding. I have tested the methods for different settings, backbones and datasets, and evaluated performance using both an annotated test set as well as visual comparisons in video.
The final result of the tests show that video object detection is a very promising area for maritime navigation, even though further testing is necessary. Tests show that video object detection methods can detect objects further away, with higher accuracy, and are less sensitive to the quality of the dataset than single-frame methods, though at a much slower speed. The tests also show that video object detection can additionally be very useful for post-analysis of video data. (Less) - Popular Abstract
- This report investigates if machine learning methods for finding navigation marks in video recorded from a ship based on data in previous video frames can be used to improve performance compared to methods that find navigation marks by looking at one image at a time.
Navigation at sea is the task of planning which route to take to get your boat from point A to point B, considering the waters, obstacles, and other boats. There are many navigation marks, such as buoys and lighthouses, that help with this. For safe navigation, it is necessary to know exactly where you are and which direction you are going. Most vessels relies on GPS for this, in addition to the navigator of the ship who keeps watch over the surrounding area. However,... (More) - This report investigates if machine learning methods for finding navigation marks in video recorded from a ship based on data in previous video frames can be used to improve performance compared to methods that find navigation marks by looking at one image at a time.
Navigation at sea is the task of planning which route to take to get your boat from point A to point B, considering the waters, obstacles, and other boats. There are many navigation marks, such as buoys and lighthouses, that help with this. For safe navigation, it is necessary to know exactly where you are and which direction you are going. Most vessels relies on GPS for this, in addition to the navigator of the ship who keeps watch over the surrounding area. However, relying on GPS can be a risk as there is a chance that the GPS signal can be lost or jammed. It is also very hard for the navigator to distinguish different types of navigation marks from distance, and thus it would be beneficial to have a system that can assist with finding what type of mark it is.
To help decrease the dependency on GPS signal Saab Kockums is developing a system that records video in real time, finds the location of navigational marks in the video, and compares the found marks to the sea chart. This information is then used to find the exact position of the ship. This solution has limitations however, one issue being that the system is not very good at detecting navigation marks that are far away or small in size. In order to become better at this, Saab Kockums wants to see if comparing information between consecutive frames can lead to higher quality detections as well as finding navigation marks that are further away.
In this report, methods for using video data to give better detections are investigated, both for running in real-time as well as analysing afterwards for creating better data for training. It was found that methods using video data can find navigation marks further away and with better accuracy than methods looking at one image at a time. However, they are much slower. It was also found that methods using video data were better at finding navigation marks in a video from a previously unseen environment.
These results imply that using video data for finding navigational objects is a promising technique, both for real-time applications and for creating new training data. Further investigation and tests are necessary to see if the video methods can become fast enough to be used in real-time on a boat. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9150834
- author
- Branzell, Oskar LU
- supervisor
- organization
- course
- FMAM05 20241
- year
- 2024
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Video Object Detection, Object Detection, Machine Learning, Maritime Navigation
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- 2024:E15
- ISSN
- 1404-6342
- other publication id
- LUTFMA-3530-2024
- language
- English
- id
- 9150834
- date added to LUP
- 2024-05-16 12:01:58
- date last changed
- 2024-05-16 12:01:58
@misc{9150834,
abstract = {{The purpose of this thesis was to investigate whether video object detection methods can outperform single-frame object detection methods in the context of maritime navigation. Object detection is a well-explored area within machine learning with many high performing published methods. Video object detection however, where information from previous video frames is considered in order to capture historical data such as movement in the detections, is a relatively recent area.
In this thesis I have researched current state-of-the-art within video object detection and applied these methods to the data provided by Saab Kockums. The video object detection methods can successfully detect navigation objects in video, even small and far away objects that single-frame object detection methods have difficulties finding. I have tested the methods for different settings, backbones and datasets, and evaluated performance using both an annotated test set as well as visual comparisons in video.
The final result of the tests show that video object detection is a very promising area for maritime navigation, even though further testing is necessary. Tests show that video object detection methods can detect objects further away, with higher accuracy, and are less sensitive to the quality of the dataset than single-frame methods, though at a much slower speed. The tests also show that video object detection can additionally be very useful for post-analysis of video data.}},
author = {{Branzell, Oskar}},
issn = {{1404-6342}},
language = {{eng}},
note = {{Student Paper}},
series = {{Master's Theses in Mathematical Sciences}},
title = {{Video Object Detection for Maritime Navigation}},
year = {{2024}},
}