Detekcia a rozpoznávanie dopravných značiek na obrázkoch pomocou sietí typu "vision transformer"

Abstract

This master’s thesis explores the possibilities of the current STA models in object detection. Its focus is on comparison of precision and efficiency of YOLOv11 detector and transformer-based neural network - RT-DETR. Experiments are realised on the traffic signs dataset which affect maximal allowed speed on roads. Evaluation is being done by the means of precision metrics, inference speed and robustness in different lighting conditions. The results of thesis provide overview whether transformer-based neural networks are closing in to the performance of YOLO detector