Master Thesis: Development of algorithms for labelrecognition

In the production halls of ArcelorMittal Gent, steel coils are transferred by manual and automatic cranes. Unfortunately, some of these transfers can go wrong: a manual crane can transfer the wrong coil or the labels, which identify the coil, can be swapped. These problems cause long delays, high costs and recurring damage for the company. To solve these kinds of problems, a solution is introduced, which consists of locating the position of the label followed by optical character recognition. It is shown that a label can be located in 82% of the cases. Implementation in the existing system of ArcelorMittal Gent shows that 64% of the located labels can be recognized successfully and that the solution can be used in a real-time environment. Finally, some techniques are discussed which can improve the result.

Download

Poster Scriptie Comprehensive proposal

Problems to be solved

  • Are there existing algorithms, tools and/or libraries that can solve the specific problem? What are the pros and contras?
  • How can the position of the label be situated?
  • What happens when multiple labels are detected?
  • How can the ID be recognized?
  • Which heuristic method can be used to determine if the ID is correct?
  • How can the algorithm be implemented in the existing system?
  • Which image processing techniques can be used?

Video (Dutch)

Presentation (Dutch)



Hey, I'm Cédric a software engineer, who's motivated to broaden his horizon and to discover and learn new methodologies. With a huge interest in web technologies, artificial intelligence and image processing, I try to understand our mountainous world a little bit better.