Primjena neuronskih mreža u ekstrinzičnoj kalibraciji robotskog in-hand stereo-vizijskog sustava

undergraduate thesis

• Title (english): Application of Neural Networks in Extrinsic Calibration of Robotic In-Hand Stereo Vision System
• Author: Lea Samaržija
• Mentor: Filip Šuligoj (mentor)
• Granter: University of Zagreb; Faculty of Mechanical Engineering and Naval Architecture; Zagreb
• Defense date and country: 2024-09-26, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Mechanical Engineering

Abstract

U ovom završnom radu obrađuje se primjena neuronskih mreža u procesu ekstrinzične kalibracije stereo-vizijskih sustava unutar robotske ruke. Sustav stereo-vizijskih kamera omogućava preciznu lokalizaciju objekata u prostoru, a kalibracija igra ključnu ulogu u osiguravanju točnosti tih mjerenja. Tradicionalne metode kalibracije često su kompleksne i dugotrajne, stoga se u ovom radu koriste neuronske mreže za automatizaciju i pojednostavljenje procesa kalibracije. U ovom radu setovi podataka napravljeni su simulacijom modela stereo kamere. Stoga ovaj rad obuhvaća pregled teorije kamera i stereo kamera, te prikazuje model stereo kamere korišten za simulaciju potrebnih setova podataka. Nakon toga, rad se fokusira na razvoj i testiranje više modela neuronskih mreža, uključujući višeslojni perceptron (MLP) i konvolucijske neuronske mreže (CNN), s ciljem poboljšanja preciznosti pozicioniranja objekata u stvarnim aplikacijama poput autonomnih vozila i robotske manipulacije.

Abstract (english)

This thesis addresses the application of neural networks in the process of extrinsic calibration of stereo vision systems within a robotic arm. The stereo vision camera system enables precise localization of objects in space, with calibration playing a crucial role in ensuring the accuracy of these measurements. Traditional calibration methods are often complex and time-consuming, therefore, this paper employs neural networks to automate and simplify the calibration process. The data sets used in this research were generated by simulating a stereo camera model. Consequently, the thesis includes an overview of camera and stereo camera theory, as well as a presentation of the stereo camera model used for data set simulation. Following this, the work focuses on the development and testing of several neural network models, including a multilayer perceptron (MLP) and convolutional neural networks (CNN), aiming to improve the accuracy of object positioning in real-world applications such as autonomous vehicles and robotic manipulation.

• Keywords: stereo-kamera; neuronska mreža; ekstrinzična kalibracija; MLP; CNN
• Keywords (english): stereo camera; neural network; extrinsic calibration; MLP; CNN
• Language: croatian
• URN:NBN: urn:nbn:hr:235:019312
• Study programme: Title: Mechanical Engineering; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka strojarstva (sveučilišni/a prvostupnik/prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka strojarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2024-09-25 09:45:01