Detekce dopravních značek ve videosekvencích

Abstract

This bachelor’s thesis focuses on the detection of traffic signs in the Czech Republic using convolutional neural networks, specifically the YOLO and Faster R-CNN architectures. For training purposes, a custom artificial dataset called CATSD was created, consisting of Czech traffic signs, and its performance was compared to the publicly available Mapillary dataset which consists of traffics signs from different countries. Testing the models across both datasets revealed a general trend of lower performance when evaluated on datasets different from those they were trained on. However, the models trained on the custom CATSD dataset achieved significantly better results in the Czech environment compared to those trained on the Mapillary dataset, suggesting their potential for practical applications.