Digitala Vetenskapliga Arkivet

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
Efficient Implementation and Upgrade of a Computer Vision System
KTH, School of Electrical Engineering and Computer Science (EECS).
2025 (English)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesisAlternative title
Effektiv implementation och uppgradering av ett computer vision-system (Swedish)
Abstract [en]

The use of computer vision systems in industrial verification processes is becoming more common due to their ability to increase efficiency and reduce errors. Setting up a computer vision system requires a lot of training data and annotating this data can be cumbersome. The industry is also advancing rapidly and systems quickly become outdated. Production workers have deemed the legacy YOLOv5-based system at Aritco inefficient and unreliable. This thesis explores the challenges and benefits of upgrading an existing computer vision system, focusing on increasing prediction accuracy and being able to perform system upgrades with reduced manual work. The thesis aims to create an efficient process that can be used to implement new and upgrade existing computer vision systems. By using the Design Science Research methodology, the goal is to create a process that by using semi-automated annotation techniques together with an iterative model retraining schedule can create a system that improves over time. Results show that the new YOLOv10 system achieves a prediction accuracy of 96%, whereas the old YOLOv5 had a prediction accuracy of only 87%. The semi-automated annotation process also reduced the time needed for image annotation by over 50%. These results render a more reliable verification system for the company while reducing the labor required to upgrade the system in the future. More generally, this study contributes to the field of computer vision by presenting a scalable and generic approach to upgrading and implementing efficient verification systems.

Abstract [sv]

Användningen av computer-vision system i industriella verifieringsprocesser blir allt vanligare på ground av dess förmåga att öka effektiviteten och minska fel i produktion. Att implementera ett computer-vision system kräver mycket träningsdata och annoteringen av denna data kan vara tidskrävande och besvärlig. Computer-vision utvecklas också snabbt, vilket leder till att modeller snabbt blir utdaterade. Det befintliga YOLOv5 systemet på Aritco har beskrivits som ineffektivt och opålitligt av personalen inom production. Denna uppsats utforskar möjligheterna och fördelarna med att uppgradera ett befintligt computer-vision system, med fokus på att förbättra effektiviteten och uppgraderingsprocessen med hjälp av semi-automatiserade annoteringstekniker. Målet med uppsatsen är att skapa en effektiv process osm kan användas både för att implementera nya samt att uppgradera befintliga computer-vision system. Genom användandet av Design Science Research metodologin syftar studied till att utveckla en process som, genom att använda semi-automatiserade annoteringstekniker tillsammans med en iterativ modellträning, kan skapa ett system som förbättras över tid. Resultaten visar att det nya YOLOv10-systemet uppnår en prediktionsgrad på 96% en förbättring från det tidigare YOLOv5 systemets grad på 87%. Den semi-automatiserade annoteringsprocessen minskade också tiden som krävdes för att annotera bilder med över 50%. Studien bidrar till computer-vision fältet genom att presentera ett skalbart och generiskt tillvägagångssätt för att implementera och uppgradera effektiva verifikationssystem.

Place, publisher, year, edition, pages
2025. , p. 37
Series
TRITA-EECS-EX ; 2025:15
Keywords [en]
Computer Vision, Object Detection, Auto-Annotation, Machine Learning, Model Upgrade, YOLO
Keywords [sv]
Datorseende, Objektigenkänning, Auto-Annotering, Maskininlärning, Mo- delluppgradering, YOLO
National Category
Computer and Information Sciences
Identifiers
URN: urn:nbn:se:kth:diva-361683OAI: oai:DiVA.org:kth-361683DiVA, id: diva2:1947303
External cooperation
Aritco Lift AB
Supervisors
Examiners
Available from: 2025-03-31 Created: 2025-03-25 Last updated: 2025-03-31Bibliographically approved

Open Access in DiVA

fulltext(4671 kB)33 downloads
File information
File name FULLTEXT02.pdfFile size 4671 kBChecksum SHA-512
fd9851916e367bfcfec6357db3eea8967852ac0c87a1666d2e6fb90fe3439526ef7ed34a81b7b6532a139862a0af30179396cec3683c2bde48bfb1df24dfaeea
Type fulltextMimetype application/pdf

By organisation
School of Electrical Engineering and Computer Science (EECS)
Computer and Information Sciences

Search outside of DiVA

GoogleGoogle Scholar
Total: 34 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 309 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