Lokalizace klíčových bodů pomocí neuronových sítí

Abstract

This bachelor thesis focuses on the task of localizing keypoints and their subsequent correspondence for estimating object pose using PnP methods with neural networks. The aim of the thesis is to research deep neural networks for keypoint localization and to implement selected approaches. In this work, primarily the models from the U-Net family are researched and implemented. This includes models that utilize the STN (Spatial Transformer Network), ensuring spatial invariance. The results of several trained model approaches are evaluated and compared using both real and synthetic images.