Pronalaženje i lociranje mobilnog manipulatora u strmom vinogradu pomoću bespilotne letjelice

master's thesis

• Title (english): Finding and locating a mobile manipulator in a steep vineyard using a UAV
• Author: Karlo Valentin Cihlar
• Mentor: Zdenko Kovačić (mentor)
• Committee member: Zdenko Kovačić (član povjerenstva)
• Committee member: Tamara Petrović (član povjerenstva)
• Committee member: Goran Vasiljević (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2022-07-14, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering

Abstract

Težak fizički rad u vinogradu može imati negativan utjecaj na ljudsko zdravlje. U okviru projekta HEKTOR razvijen je sustav za nadzor i obavljanje posla u vinogradu pomoću autonomnih robota - bespilotne letjelice i mobilnog manipulatora. Kako bi se locirao mobilni robot u slučaju kvara ili zaglavljivanja, u ovom je radu razvijen algoritam za detekciju i pronalazak mobilnog robota VIV-a u strmom vinogradu. Testirane su raličite metode detekcije te je na kraju odabrana detekcija pomoću neuronske mreže. Napravljen je skup podataka sa slikama VIV-a te je istrenirana YOLOv5 duboka konvolucijska neuronska mreža za detekciju na tom podatkovnom skupu. Mreža daje vrlo dobre rezultate, ispravno detektirajući robota u većini slučajeva s visokom razinom uvjerenja (>=0.65) na dotad neviđenim podacima. Na taj način omogućava jednostavan pronalazak moblinog robota u vinogradu. U budućnosti se može implementirati efikasnija metoda pretraživanja vinograda te se može proširiti podatkovni skup kako bi neuronska mreže bila još robusnija.

Abstract (english)

Hard physical work in the vineyard can have a negative impact for human health. Within the HEKTOR project, a system was developed for monitoring and performing work in the vineyard with the help of autonomous robots - unmanned aerial vehicle and a mobile manipulator. In order to locate the mobile robot in case of failure or jamming, an algorithm for detecting and finding the VIV mobile robot in a steep vineyard was developed in this thesis. Different detection methods were tested and finally neural network detection was selected. A dataset with VIV images was created and a YOLOv5 deep convolutional neural network was trained for detection on that data set. The network gives very good results, correctly detecting the robot in most cases with a high level of confidence (>=0.65) on unseen data. This makes it easy to find the mobile robot in the vineyard. In the future, a more efficient method of searching vineyards can be implemented and the dataset can be expanded to make neural network even more robust.

• Keywords: HEKTOR; VIV; detekcija; neuronska mreža; vinograd; bespilotna letjelica
• Keywords (english): HEKTOR; VIV; object detection; neural network; vineyard; UAV
• Language: english
• URN:NBN: urn:nbn:hr:168:822003
• 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 14:41:03