Detekcija cesta koristeći LiDAR podatke dobivene zračnim snimanjem RH

master's thesis

• Title (english): Road detection using LiDAR data acquired from aerial surveys of the Republic of Croatia
• Author: Kate Gojak
• Mentor: Mario Miler (mentor); MBZ: 296874
• Mentor: Ivan Brkić (pomoć pri izradi, neposredni voditelj)
• Committee member: Luka Rumora (predsjednik povjerenstva)
• Committee member: Mario Miler (član povjerenstva)
• Committee member: Dino Dobrinić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Geodesy; (Institute of Geomatics); (Chair of Geoinformatics); Zagreb
• Defense date and country: 2024-06-21, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Geodesy; Geomatics

Abstract

Cilj ovog rada je primjenom dostupnih ulaznih slojeva – digitalnog ortofoto (DOF) rastera, LiDAR (engl. Light Detection and Ranging) podataka te vektorskog sloja cesta preuzetog s OpenStreetMap-a detektirati i izdvojiti cestovni sloj u urbanom području grada Karlovca. Korišteni softver, eCognition Developer, nudi napredne funkcije za segmentiranje i klasificiranje slika. Za segmentaciju se koristi multirezolucijska segmentacija, pristup temeljen na analizi objekata (engl. Object-based Image Analysis - OBIA). Klasifikacija je nadzirana, što uključuje ručno odabiranje reprezentativnih uzoraka za svaku klasu. Proces klasifikacije koristi srednje vrijednosti ulaznih slojeva i algoritam klasifikatora najbližeg susjeda (engl. k-nearest neighbors - kNN) kako bi dodijelio preostale segmente jednoj od dvije definirane klase: cesta i ne-cesta. Analizom utjecaja pojedinih ulaznih slojeva na rezultate klasifikacije, cilj je dublje razumjeti izazove u izdvajanja cesta u urbanom okruženju te raspraviti moguća rješenja. Kombinacija spektralnih kanala, visinskih podataka i vektorskog sloja cesta pokazala se ključnom za poboljšanje točnosti klasifikacije. Međutim, najveći izazov ostaje pravilna klasifikacija ostalih asfaltiranih površina, poput parkirališta i pločnika, koje su spektralno slične cestama. Iako su neki segmenti ovih površina uspješno eliminirani, varijacije u rezultatima i dalje su prisutne. Ovaj rad pruža temelj za buduća istraživanja i unaprjeđenja u području segmentacije i klasifikacije urbanih cesta, ukazujući na važnost integracije različitih izvora podataka i naprednih algoritama za analizu slika.

Abstract (english)

The aim of this study is to detect and extract the road surfaces in the urban area of Karlovac by applying available input layers – digital orthophoto (DOF) raster, LiDAR data, and a line vector layer of roads obtained from OpenStreetMap. eCognition Developer software was used. It offers advanced functions for image segmentation and classification. Multiresolution segmentation, an approach based on object-based image analysis (OBIA), is used for segmentation. The classification is supervised, which includes manually selecting representative samples for each class. The classification process utilizes the mean values of input layers and the k-nearest neighbor (kNN) classifier algorithm to assign the remaining segments to one of two defined classes: road and non-road. By analyzing the impact of individual input layers on classification results, the goal is to gain a deeper understanding of the challenges in road extraction in urban environments and to discuss possible solutions. The combination of spectral bands, elevation data, and the line vector road layer proved to be crucial for improving classification accuracy. However, the greatest challenge remains the accurate classification of other asphalt surfaces, such as parking lots and sidewalks, which are spectrally similar to roads. Although some segments of these surfaces were successfully eliminated, variations in results are still present. This study provides a base for future research and advancements in the field of urban road segmentation and classification, highlighting the importance of integrating different data sources and advanced image analysis algorithms.

• Keywords: ceste; eCognition; klasifikacija; LiDAR; OBIA; segmentacija
• Keywords (english): roads; eCognition; classification; LiDAR; OBIA; segmentation
• Language: croatian
• URN:NBN: urn:nbn:hr:256:363009
• 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: Embargoed access; Embargo expiration date: 2034-06-01
• Created on: 2024-06-21 11:45:55