Rozpoznávání objektů a jejich polohy na základě 3D modelu

Abstract

The thesis deals with the recognition of a larger number of objects and the estimation of their poses based on their 3D model. Recognition takes place from 3D maps obtained from various depth sensors. The goal of this work is to implement software for recognition using state of the art methods included in the Point Cloud Library and try to improve any step of the recognition chain. Using a neural network, I decided to modify the method for finding corresponding pairs of key points. I describe the motivation for the use of a neural network, its advantages and dis-advantages. Finally, some of its parameters are experimentally tested. There is also a comparison of the neural network with the classical method.