Primerjava metod za detekcijo puščic pri klasičnem pikadu

ZGONC, MATIC (author), Batagelj, Borut (mentor)

V diplomski nalogi smo preverili, če se klasične metode računalniškega vida lahko primerjajo z modernimi pristopi, ki temeljijo na strojnem učenju in nevronskih mrežah. Primerjavo smo izvedli na problemu detekcije puščic pri klasičnem pikadu. Cilj je bil razviti sistem z visoko natančnostjo, ki bo enostaven za uporabo in cenovno dostopen. S pomočjo metod računalniškega vida smo najprej razvili klasični sistem, kjer smo opisali celoten postopek, od kalibracije do detekcije puščic. Sledil je razvoj sistemov, ki temeljijo na strojnem učenju. Uporabili smo različne arhitekture, da smo ugotovili, katera izmed njih prinaša najboljše rezultate. Po pridobitvi vseh rezultatov smo preverili še časovno zahtevnost posameznih pristopov. Sledila je končna primerjava sistemov glede na uspešnost in zmogljivost, ter cenovna primerjava razvitega sistema z obstoječimi rešitvami.

• Language: Slovenian
• Keywords: računalniški vid, OpenCV, pikado, YOLO, semantična segmentacija, instančna segmentacija
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2022
• PID: 20.500.12556/RUL-140925
• COBISS.SI-ID: 124605187
• Publication date in RUL: 21.09.2022

Secondary language

• Language: English
• Title: Comparison of arrow detection methods for classic darts

Abstract

In the thessis we compared classic methods of computer vision with modern approaches based on machine learning and neural networks. Comparison was made on problem of detecting arrows at steel darts. First we developed system based on classic methods of computer vision and described complete process from calibration to score prediction. Next we developed systems based on machine learning. We used different architectures to find out which give us the best results. After we gathered all results we checked time complexity on all developed systems. At the end we compared systems by detection rate, performance and checked price difference of our system with existing solutions.

• Keywords: computer vision, OpenCV, darts, YOLO, semantic segmentation, instance segmentation