Automatska detekcija stanja prostorne baze podataka vertikalne signalizacije koristeći metode strojnog učenja i georeferencirani video

master's thesis

• Title (english): Automatic Detection of the Condition of the Traffic Sign Spatial Database Using Machine Learning Methods and Georeferenced Video
• Author: Luka Meštrić
• Mentor: Mario Miler (mentor)
• Mentor: Ivan Brkić (pomoć pri izradi, neposredni voditelj)
• Committee member: Mario Miler (predsjednik povjerenstva)
• Committee member: Luka Rumora (član povjerenstva)
• Committee member: Dino Dobrinić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Geodesy; Zagreb
• Defense date and country: 2024-09-06, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Geodesy

Abstract

Vertikalna signalizacija svakodnevno je prisutna u našim životima, vodeći nas kroz
sigurno kretanje po cestama. Ona pruža bitne informacije poput ograničenja brzine,
upozorenja na opasnosti i smjerova kretanja, osiguravajući tako sigurnost i red u prometu.
Bez vertikalne signalizacije, odnosno prometnih znakova promet bi bio rizičan.
Održavanje ažurne prostorne baze vertikalne cestovne signalizacije to jest prometnih
znakova je zahtjevan zadataka za bilo kojeg održavatelja cesta.
U ovom diplomskom radu istraživala se mogućnost kreiranja mehanizma koji automatski
detektira stanje prostorne baze podataka vertikalne signalizacije čime bi se olakšalo
održavanje iste. Drugim riječima cilj je obaviti inventuru određene baze podataka
prometnih znakova koristeći suvremene automatizirane metode. Istraživanje se provelo
na podacima koji obuhvaćaju područje Brodsko-posavske županije. Podaci se sastoje od
prostorne baze podataka vertikalne signalizacije i georeferenciranog videa. Istraživanje
je obavljeno s pomoću YOLO algoritma, Python programskog jezika i QGIS-a.
Dodatno je provedena analiza modela za predikciju prometnih znakova koji je dobiven s
pomoću YOLOv8 algoritma. Cilj ove analize bio je ocijeniti model, njegovu preciznost i
točnost kako vi se vidjelo je li model dovoljno dobar za korištenje u ovom diplomskom
radu, ali potencijalno u nekim drugim istraživanjima.
Dobiveni rezultati dokazali su da se korištenjem suvremenih metoda kao što je strojno
učenje može olakšati održavanje prostorne baze podataka vertikalne signalizacije.
Drugim riječima može se uspješno obaviti automatizirana inventura prometnih znakova
za neko odabrano područje. Uz uspješno obavljanje cilja zaključeno je da prostora za
napredak u automatskoj detekciji stanja prostorne baze podataka vertikalne signalizacije
ima u vidu poboljšanja modela predikcija i korištenja videa koji snima oba smjera ceste.

Abstract (english)

Vertical signaling is present in our lives every day, guiding us through safe movement on
the roads. It provides essential informations such as speed limits, warnings of dangers and
directions of movement, thus ensuring safety and order in traffic. Without vertical
signaling, in other words traffic signs, traffic would be risky. Maintaining an up-to-date
spatial base of vertical road signage, i.e. traffic signs, is a demanding task for any road
maintainer.
In this thesis, the possibility of creating a mechanism that automatically detects the state
of the spatial database of vertical signage was studied, which would facilitate its
maintenance. In other words, the goal is to perform an inventory of a specific traffic sign
database using modern automated methods. The research was conducted on data covering
the area of Brodsko-posavska županija. The data consists of a spatial database of vertical
signage and georeferenced video. The research was carried out using the YOLO
algorithm, Python programming language and QGIS.
In addition, an analysis of the traffic sign prediction model obtained using the YOLOv8
algorithm was performed. The goal of this analysis was to evaluate the model, its
precision and accuracy as you could see if the model is good enough to be used in this
thesis, but potentially in some other research.
The obtained results proved that the maintenance of the spatial database of vertical
signaling can be facilitated by using modern methods such as machine learning. In other
words, an automated traffic sign inventory can be successfully performed for a selected
area. With the successful completion of the goal, it was concluded that there is room for
improvement in the automatic detection of the state of the spatial database of vertical
signaling in terms of improving the prediction model and the use of video that records
both directions of the road.

• Keywords: Vertikalna signalizacija; prometni znakovi; Yolo algoritam; Python
• Keywords (english): Vertical signaling; traffic signs; Yolo algorithm; Python
• Language: croatian
• URN:NBN: urn:nbn:hr:256:366938
• Study programme: Title: Geodesy and Geoinformatics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka geodezije i geoinformatike (magistar/magistra inženjer/inženjerka geodezije i geoinformatike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2024-09-12 10:59:39