umu.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Improvement Methods for Assignment Problem in Multi-target Tracking
Umeå University, Faculty of Science and Technology, Department of Physics.
2019 (English)Independent thesis Advanced level (professional degree), 20 credits / 30 HE creditsStudent thesis
Abstract [en]

Using computer vision to track moving objects is an important but complicated task. In particular, tracking human motion is difficult, since the motion is often complex with motion in three dimensions making it hard to track and model. Accurate position tracking is, however, important in order to estimate the performance during strength and agility training since the position data can be used for several measures. For example, from the position the maximum velocity in a sprint and force development weightlifting can be assessed. In this project, the aim was to implement a software capable of performing multi-target tracking during human motion. We evaluated two specific algorithms optimized for this: the Hungarian algorithm and the Kalman filter, in which the latter was expanded into the interactive multiple model (IMM) algorithm. We implemented these algorithms and verified the performance using a simple multi-target tracking scenario. The results showed that these methods are fast and that they represent a good foundation for displaying measurement results in real-time. Furthermore, they provide a basis for handling more advanced problems related to multitarget tracking, such as object occlusion and collisions. We validated the software output using reference data obtained from a 1080 sprint and 1080 quantum device during a few exercises. In general, the software produces similar output as the reference data on a large scale. On a more local scale, however, the residuals between the camera and 1080 devices vary depending on the exercise. For example, in a squat jump exercise, a typical residual in position is around 5 cm. Whereas, for short-distance sprint, the residuals can be up to 40 cm, although these are partly due to differences in the measuring techniques themselves. The results imply that the algorithms can still be modified to achieve higher accuracy. As the quality of the output seems to vary depending on the tracked motion, it is possible that unique sub-routines may need to be tailored for each exercise. Still, the implementations made during the project can be used in simple measurement scenarios, and serve as a good basis for future work.

Place, publisher, year, edition, pages
2019.
National Category
Signal Processing
Identifiers
URN: urn:nbn:se:umu:diva-156107OAI: oai:DiVA.org:umu-156107DiVA, id: diva2:1286151
External cooperation
Photon Sports Technologies AB
Subject / course
Examensarbete i teknisk fysik
Educational program
Master of Science Programme in Engineering Physics
Supervisors
Examiners
Available from: 2019-02-12 Created: 2019-02-06 Last updated: 2019-02-12Bibliographically approved

Open Access in DiVA

No full text in DiVA

Search in DiVA

By author/editor
Nygård, Marko
By organisation
Department of Physics
Signal Processing

Search outside of DiVA

GoogleGoogle Scholar

urn-nbn

Altmetric score

urn-nbn
Total: 83 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf