Detekce přechodů pomocí technik hlubokého učení a zpracování obrazu

Abstract

Bachelor thesis focuses on the usage of Convolutional neural networks to solve the task of automatic detection of crosswalks from aerial imagery from the cities of Ostrava and Hradec Králové. This thesis includes research of models, tools, and technologies, which are often used for object detection tasks. During the experimentation, the most suitable tools were used along with the most popular models. Model with the best results was then deployed on different area within these cities and used for inference. This thesis can be used as a methodology to solve problems dealing with automatic detection of crosswalks in aerial imagery.