Neural Network-Based License Plate Recognition
Shlenchak, Olga (2025)
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2025051210754
https://urn.fi/URN:NBN:fi:amk-2025051210754
Tiivistelmä
The purpose of this bachelor's thesis was to create a system for recognizing state registration numbers of cars.
The subject area was analyzed, and existing systems with similar functionality were reviewed. An informational-logic model of the system was developed: UML diagrams were created, algorithms were designed, and system architecture was documented using the Draw.io web application.
An automated system recognizing license plates (ASRN) was implemented using Python and Windows Forms in the PyCharm 2022 development environment, targeting Windows 10 and newer operating systems.
The ASRN was trained and validated using a dataset of license plate images. Performance evaluation on a test set of 1,000 photo frames showed an average recognition time of 93 milliseconds and an accuracy rate of 83%. In terms of accuracy, the developed system performs on par with commercial LPR solutions such as CVS Auto and Auto Trassir. The evaluation also included speed and robustness under challenging conditions such as low light and weather interference. While results demonstrate high efficiency, further training and optimization are planned to enhance overall accuracy and reliability.
The subject area was analyzed, and existing systems with similar functionality were reviewed. An informational-logic model of the system was developed: UML diagrams were created, algorithms were designed, and system architecture was documented using the Draw.io web application.
An automated system recognizing license plates (ASRN) was implemented using Python and Windows Forms in the PyCharm 2022 development environment, targeting Windows 10 and newer operating systems.
The ASRN was trained and validated using a dataset of license plate images. Performance evaluation on a test set of 1,000 photo frames showed an average recognition time of 93 milliseconds and an accuracy rate of 83%. In terms of accuracy, the developed system performs on par with commercial LPR solutions such as CVS Auto and Auto Trassir. The evaluation also included speed and robustness under challenging conditions such as low light and weather interference. While results demonstrate high efficiency, further training and optimization are planned to enhance overall accuracy and reliability.