Advanced

Tracking the Motion of Box Jellyfish

Kjellberg, Tobias LU (2014) In Master's Theses in Mathematical Sciences FMA820 20131
Mathematics (Faculty of Technology) and Numerical Analysis
Abstract
This master’s thesis investigates the possibilities of detecting the rhopalia of box jellyfishes, i.e. the eyes of the box jellyfish. Each box jellyfish has four rhopalia and each of the rhopalium consists of six light sensing eyes which they use to interpret the surroundings. In this project we have worked with fifteen film sequences all consisting of different box jellyfishes and recorded under different light settings. The framework for detecting the rhopalia is divided into three parts, in order; detection, clustering and tracking. The input in the detection step is the grayscale image of the box jellyfish and in the output possible rhopalia are marked as detections. These detections are then sent into the clustering step which filters... (More)
This master’s thesis investigates the possibilities of detecting the rhopalia of box jellyfishes, i.e. the eyes of the box jellyfish. Each box jellyfish has four rhopalia and each of the rhopalium consists of six light sensing eyes which they use to interpret the surroundings. In this project we have worked with fifteen film sequences all consisting of different box jellyfishes and recorded under different light settings. The framework for detecting the rhopalia is divided into three parts, in order; detection, clustering and tracking. The input in the detection step is the grayscale image of the box jellyfish and in the output possible rhopalia are marked as detections. These detections are then sent into the clustering step which filters out the noise in the picture and saves clusters of detections. The reason for saving these clusters is because a rhopalium appears as a dark disc with a radius of around 13 pixels and thus should produce many detections. In the final step, four clusters are selected from all clusters as the correct rhopalia. The choice has been made to focus on the detection step in this thesis, leaving the cluster and tracking step with only one algorithm each. A combination of a detection method, a clustering- and a tracking algorithm is called a system.
To be able to detect a rhopalium, a set of data points, a pattern, is used to compare values (light intensity) in order to capture the appearance of a rhopalium. The more data points the longer the execution time so this needs to be done with minimal amount of data points but still enough to capture the visual aspect of the rhopalium.
The best performing systems have patterns of a small disc inside the rhopalium and a larger circle just outside. The values of the pixels outside should be greater (brighter) than the ones inside. But because of some artefacts and noise in the pictures not all need to be darker, instead, a threshold is used, e.g 78 % of the pixel values outside should be greater than the ones inside. This results in a accuracy of 98 %. (Less)
Please use this url to cite or link to this publication:
author
Kjellberg, Tobias LU
supervisor
organization
course
FMA820 20131
year
type
H2 - Master's Degree (Two Years)
subject
keywords
image analysis, tracking
publication/series
Master's Theses in Mathematical Sciences
report number
LUTFMA-3260-2014
ISSN
1404-6342
other publication id
2014:E26
language
English
id
4463255
date added to LUP
2014-07-04 17:26:55
date last changed
2014-07-04 17:26:55
@misc{4463255,
  abstract     = {This master’s thesis investigates the possibilities of detecting the rhopalia of box jellyfishes, i.e. the eyes of the box jellyfish. Each box jellyfish has four rhopalia and each of the rhopalium consists of six light sensing eyes which they use to interpret the surroundings. In this project we have worked with fifteen film sequences all consisting of different box jellyfishes and recorded under different light settings. The framework for detecting the rhopalia is divided into three parts, in order; detection, clustering and tracking. The input in the detection step is the grayscale image of the box jellyfish and in the output possible rhopalia are marked as detections. These detections are then sent into the clustering step which filters out the noise in the picture and saves clusters of detections. The reason for saving these clusters is because a rhopalium appears as a dark disc with a radius of around 13 pixels and thus should produce many detections. In the final step, four clusters are selected from all clusters as the correct rhopalia. The choice has been made to focus on the detection step in this thesis, leaving the cluster and tracking step with only one algorithm each. A combination of a detection method, a clustering- and a tracking algorithm is called a system.
To be able to detect a rhopalium, a set of data points, a pattern, is used to compare values (light intensity) in order to capture the appearance of a rhopalium. The more data points the longer the execution time so this needs to be done with minimal amount of data points but still enough to capture the visual aspect of the rhopalium.
The best performing systems have patterns of a small disc inside the rhopalium and a larger circle just outside. The values of the pixels outside should be greater (brighter) than the ones inside. But because of some artefacts and noise in the pictures not all need to be darker, instead, a threshold is used, e.g 78 % of the pixel values outside should be greater than the ones inside. This results in a accuracy of 98 %.},
  author       = {Kjellberg, Tobias},
  issn         = {1404-6342},
  keyword      = {image analysis,tracking},
  language     = {eng},
  note         = {Student Paper},
  series       = {Master's Theses in Mathematical Sciences},
  title        = {Tracking the Motion of Box Jellyfish},
  year         = {2014},
}