Design and implementation of a system for automatic licence plate recognition using computer vision

This thesis deals with the design and comparative analysis of architectures for an automatic license plate recognition (ALPR) system using computer vision technologies. The aim of the thesis was to design an optimal, scalable, and efficient solution for recognizing Czech license plates and to verify its key metrics. The theoretical part focuses on deep neural networks, in particular the YOLO (You Only Look Once) architecture for object detection, and optical character recognition (OCR) methods. The practical part presents a comparative study of three pilot architectures, including approaches based on pure OCR and a hybrid model that combines the detection capabilities of the YOLOv11 model and the EasyOCR recognition algorithm. These pilot implementations were thoroughly tested on a unified test dataset, quantifying critical detection and OCR metrics. The work does not implement or provide instructions for implementing the final system for real-world deployment due to time, financial, and personnel constraints arising from the limitations of the academic environment, but it does provide a robust, data-driven solution design and detailed recommendations for further development in practical deployment and optimization of the selected architecture.