Usporedba modela računalnog vida za detekciju objekata na sklopovlju s niskim performansama

master's thesis

• Title (english): Comparison of computer vision models for object detection using low-power hardware
• Author: Stipan Ljudevit Drmić
• Mentor: Petar Mostarac (mentor)
• Committee member: Petar Mostarac (predsjednik povjerenstva)
• Committee member: Hrvoje Hegeduš (član povjerenstva)
• Committee member: Alan Šala (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2022-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering

Abstract

Prilikom izvođenja algoritama računalnog vida za potrebe detekcije objekata pretežito se koristi sklopovlje s visokim procesorskim performansama.
Pritom, ako se isti koriste u stvarnom vremenu na sklopovlju s niskim performansama javlja se problem nedostatka procesorske snage.
Kao rješenje nameće se upotreba algoritama zasnovanih na jednom prolasku kroz sliku (engl. singl shot detection).
U ovome radu napravit će se usporedba različitih modela računalnog vida za detekciju objekata na konkretnom ugradbenom sklopu niske potrošnje.
Rezultati rada će se koristiti za odabir optimalnog modela s obzirom na parametre poput primjerice potrošnje energije sklopovlja, brzine obrade podataka te preciznost odabranog modela.
Svu potrebnu literaturu osigurat će središnja knjižnica fakulteta te Zavod za osnove elektrotehnike i električna mjerenja.

Abstract (english)

When performing computer vision algorithms, high-performance hardware is predominantly used for object detection purposes.
In addition, if they are used in real time on low-performance hardware, there is a problem of lack of processing power.
As a solution, the use of algorithms based on single shot detection is imposed.
In this paper, a comparison of different computer vision models for object detection on a specific low - consumption installation will be made.
The results of the work will be used to select the optimal model with regard to parameters such as hardware power consumption, data processing speed and accuracy of the selected model.
All the necessary literature will be provided by the central library of the faculty and the Department of Fundamentals of Electrical Engineering and Electrical Measurements.

• Keywords: Računalni vid; detekcija objekata; ugradbeno sklopovlje niskih performansi; algoritmi za detekciju objekata
• Keywords (english): Computer vision; object detection; low performance embedded hardware; object detection algorithms
• Language: croatian
• URN:NBN: urn:nbn:hr:168:484376
• 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 12:09:01