Semantička segmentacija rupa na cesti za autonomni vizualni pregled bespilotnom letjelicom

master's thesis

• Title (english): Semantic segmentation of road potholes for autonomous visual inspection with unmanned aerial vehicle
• Author: Filip Vuksanović
• Mentor: Stjepan Bogdan (mentor)
• Committee member: Stjepan Bogdan (predsjednik povjerenstva)
• Committee member: Zdenko Kovačić (član povjerenstva)
• Committee member: Goran Vasiljević (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-09, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering

Abstract

Bespilotne letjelice se posljednjih godina sve više koriste u svrhu inspekcije različitih infrastruktura. Ceste su jedan od primjera infrastrukture kojem je potreban redoviti pregled jer se svakodnevno pojavljuju nova oštećenja na cestama. Trenutne metode vremenski drugo traju i skupocjene su jer su ljudi ti koji izvršavaju pregled. Kako bi se riješio taj problem, u ovom diplomskom radu je predložena metoda cestovnih pregleda u kojoj se koriste podaci sa stereo kamere i senzora postavljenih na bespilotnoj letjelici. Prvo se pomoću segmentacije instanci, za koju se koristi model neuronski mreža poznat kao Mask R-CNN, otkrivaju rupe na cesti. Zatim se pozicija rupa u globalnom koordinatnom sustavu izračunava na temelju senzorskih očitanja s letjelice i pozicije letjelice u prostoru. Na kraju se očitani rezultati grupiraju aglomerativnom metodom grupiranja te se za svaku grupu računa centroid koji predstavlja poziciju rupe u globalnom sustavu. Predložena metoda je testirana u Gazebo simulatoru s letjelicom koja na
sebi ima stereo kameru. Eksperimentalni rezultati su pokazali kako se korištenjem ove metode mogu detektirati rupe te izračunati njihove pozicije u globalnom koordinatnom sustavu.

Abstract (english)

Unmanned aerial vehicles (UAVs) have been widely used for infrastructure inspections in recent years. Roads are one of those infrastructures that need constant inspection as there are new road damages every day. Current methods of road inspections
are time-consuming and costly as they are usually inspected manually. To address this
problem, this thesis proposes a process for pothole inspection based on stereo camera
and UAV’s inertial measurement units (IMUs) sensors. First, the instance segmentation
method based on Mask R-CNN was used for pothole detection. Then, the positions of
the potholes were calculated based on the UAV’s IMUu sensors and the position of
the UAV in the global coordinate system. Finally, the agglomerative clustering method
was used to cluster the calculated potholes’ positions. The centroid of each cluster represents the pothole’s coordinate in the global coordinate system. The proposed process
was implemented in the Gazebo simulator, in which the UAV model has a mounted
stereo camera. The experimental results show that this process is capable of detecting
potholes and calculating the potholes’ coordinates.

• Keywords: Bespilotna letjelica; segmentacija instanci; Mask R-CNN; grupiranje
• Keywords (english): Unmanned aerial vehicles; instance segmentation; Mask R-CNN; cluster
• Language: english
• URN:NBN: urn:nbn:hr:168:173299
• Study programme: Title: Electrical Engineering and Information Technology; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka elektrotehnike i informacijske tehnologije (magistar/magistra inženjer/inženjerka elektrotehnike i informacijske tehnologije)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 14:16:54