Razvoj aplikacije za detekciju i klasifikaciju sudionika u prometu iz video signala korištenjem modela dubokog učenja

undergraduate thesis

• Title (english): Development of an application for detection and classification of traffic from a video signal using deep learning models
• Author: Fran Sipić
• Mentor: Marko Horvat (mentor)
• Committee member: Marko Horvat (predsjednik povjerenstva)
• Committee member: Juraj Dončević (član povjerenstva)
• Committee member: Kristina Ćaleta (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-02, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Cilj ovog završnog rada bio je razviti i evaluirati aplikaciju za detekciju i klasifikaciju sudionika u prometu koristeći naprednu verziju YOLO (You Only Look Once) arhitekture, specifično YOLOv8. Model je treniran na skupu podataka koji uključuje različite sudionike prometa, poput pješaka, automobila, kamiona, autobusa i motocikala. Unatoč težnji za visokim performansama, model je pokazao manje zadovoljavajuće rezultate u usporedbi s referentnim znanstvenim radovima. U radu se analiziraju potencijalni razlozi za ove rezultate, uključujući ograničenja kvalitete i kvantitete trening podataka, tehnička ograničenja platforme Kaggle na kojoj je model treniran, te nedostatak dovoljne augmentacije podataka. Nadalje, razmatraju se strategije za poboljšanje performansi modela, kao što su proširenje i diversifikacija trening skupa podataka, produženo vrijeme treniranja i naprednije tehnike predobrade i augmentacije. Rad daje uvid u izazove i mogućnosti razvoja točnih i pouzdanih sustava za automatsku detekciju i klasifikaciju sudionika u prometu, što je ključno za implementaciju u stvarnim aplikacijama poput autonomne vožnje i inteligentnih prometnih sustava.

Abstract (english)

This thesis's objective was to develop and evaluate an application for detecting and classifying traffic participants using an advanced version of the YOLO (You Only Look Once) architecture, specifically YOLOv8. The model was trained on a dataset including various traffic participants such as pedestrians, cars, trucks, buses, and motorcycles. Despite aiming for high performance, the model demonstrated less satisfactory results compared to reference scientific studies. This thesis analyzes potential reasons for these outcomes, including limitations in the quality and quantity of training data, technical constraints of the Kaggle platform where the model was trained, and a lack of sufficient data augmentation. Furthermore, strategies for improving model performance are considered, such as expanding and diversifying the training data set, extending training time, and employing more advanced preprocessing and augmentation techniques. This work provides insights into the challenges and opportunities in developing accurate and reliable systems for the automatic detection and classification of traffic participants, which is crucial for implementation in real-world applications such as autonomous driving and intelligent traffic systems.

• Keywords: detekcija sudionika u prometu; YOLOv8; duboko učenje; klasifikacija objekata; računalni vid; augmentacija podataka; autonomna vožnja; inteligentni prometni sustavi
• Keywords (english): traffic participant detection; YOLOv8; deep learning; object classification; computer vision; data augmentation; autonomous driving; intelligent traffic systems
• Language: croatian
• URN:NBN: urn:nbn:hr:168:577287
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2024-10-12 22:05:56