Race number recognition from digital images

OData support
Supervisor:
Prekopcsák Zoltán
Department of Telecommunications and Media Informatics

During any sporting event, there can be a vast number of photographs taken. It is of great practical value to be able to list images according to the competitors appearing on them. In order to do so, it is necessary to determine which competitors are visible in each picture. Automating this process instead of doing it manually saves significant time and effort.

The problem I tacked was to create such a system that identifies competitors in images by their race numbers. My system creates a list of visible race numbers for each image. With the help of these lists, we can select the images showing any given competitor. Additionally, my application measures the accuracy of the race number recognition task and the time required for it.

I solve the problem of race number recognition in two steps. My system first locates the race numbers in the images. Secondly, it recognizes the texts belonging to the race numbers in the pictures obtained in the previous step. When given a list of the actual race numbers appearing in each photo, the recognition rate achieved by the system is calculated.

In my solution I developed and embed into my system a method for detecting and recognizing texts in scene images. My system is not a general-purpose text recognition tool, text recognition and text-processing settings are optimized for race number recognition in images.

I measured the effectiveness of the system by processing 50 images. The result is quite good, with the system achieving a recognition rate of more than 80% of the race numbers being correctly identified. The average time required for processing an image is about 67 seconds.

Downloads

Please sign in to download the files of this thesis.