Detekcija objekata na slikama temeljena na konvolucijskoj neuronskoj mreži

undergraduate thesis

• Title (english): Detection of objects in images based on a convolutional neural network
• Author: Andrea Caplić
• Mentor: Tomislav Stipančić (mentor)
• Granter: University of Zagreb; Faculty of Mechanical Engineering and Naval Architecture; Zagreb
• Defense date and country: 2022-09-28, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Mechanical Engineering

Abstract

Detekcija objekata je aktualna tehnologija povezana s računalnim vidom kojoj su cilj identifikacija i klasifikacija objekata te lociranje njihovih položaja u digitalnim slikama i videozapisima. Ima primjenu u mnogim poljima svakodnevnog života kao što su prepoznavanje lica na nadzornim videokamerama i samovozeći automobili. Čovjek nakon kratkog pogleda na sliku odmah može prepoznati koji su objekti na slici, kako su pozicionirani i u kakvom su međusobnom odnosu s ostalim objektima. Čovjekov vid je brz i precizan, a to nam omogućuje obavljanje složenih zadataka poput vožnje automobila. Brzi algoritmi za otkrivanje objekata omogućavaju i automobilima autonomnu vožnju koja trenutno postoji, no i dalje ima puno ograničenja zbog kojih ta vožnja nije moguća u svim prilikama. Postoji nekoliko metoda dubokog učenja koje se temelje na konvolucijskim neuronskim mrežama, a za potrebe ovog rada korištena je YOLOv3 metoda. U ovom radu su predstavljeni unaprijed trenirani model koji već posjeduje velik broj klasa koje prepoznaje, a potom dodatno trenirani model kako bi mogao prepoznati neke nove objekte.

Abstract (english)

Object detection is a current technology related to computer vision that aims to identify and classify objects and locate their positions in digital images and videos. It has applications in many fields of everyday life such as facial recognition on surveillance cameras and self-driving cars. After a quick glance at the picture, a person can immediately recognize which objects are in the picture, how they are positioned and what kind of relationship they have with other objects. Human vision is fast and precise and this allows us to perform complex tasks such as driving a car. Fast algorithms for detecting objects enable cars to drive autonomously, which currently exists, but there are still many limitations that do not make such driving possible in all situations. There are several deep learning methods based on convolutional neural networks but the YOLOv3 method was used for the purpose of this paper. This paper presents a pre-trained model that already has a large number of classes it recognizes and additionally trained model in order to be able to recognize some new objects.

• Keywords: YOLOv3; strojno učenje; duboko učenje; konvolucijske neuronske mreže; Keras
• Keywords (english): YOLOv3; machine learning; deep learning; convolutional neural networks; Keras
• Language: croatian
• URN:NBN: urn:nbn:hr:235:797275
• 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: 2022-09-24 13:41:28