Praćenje više pokretnih objekata temeljeno na algoritmima detekcije

master's thesis

• Title (english): Multi-object tracking based on tracking-by-detection paradigm
• Author: Domagoj Šimag
• 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

U radu je obrađeno praćenje više pokretnih objekata temeljeno na metodi praćenja pomoću detekcije. Korištene su Tracktor i Tracktor++ metode, te je radi poboljšanja razvijena Tracktor+++ metoda. Da bi se određeni objekt na snimci uspješno detektirao i pratio korištena je, zbog postignutih rezultata u tom području, konvolucijska neuronska mreža. Inačice Tracktor metode koriste Faster R-CNN, Feature Pyramid Networks, Region Proposal Network i Non-Maximum Suppression algoritam i arhitekture za pronalaženje i praćenje objekata. Implementirana su četiri matematička modela za predviđanje trajektorije kretanja objekata kroz sekvence. To su otežani pomični prosjek, alpha - beta filtar, alpha - beta - gamma filtar i Kalmanov filtar. Dodatno je analizirano kako promjena intervala detekcije utječe na performanse različitih inačica Tracktor metoda.

Abstract (english)

The subject of this master thesis is multi-object tracking based on tracking-by-detection paradigm. Tracktor and Tracktor++ methods are used and, in order to improve its performance, a Tracktor+++ method is developed. In order for a method to successfully detect and track a certain object, a convolutional neural network is used. All three types of Tracktor methods use Faster R-CNN, Feature Pyramid Networks, Region Proposal Network i Non-Maximum Suppression algorithm and architectures to detect and track objects. Four mathematical models are implemented to estimate object's trajectory throughout sequences. These are weighted moving average, alpha - beta filter, alpha - beta - gamma filter and Kalman filter. Additionally, a test is made to check how a change in detection interval affects Tracktor's performance.

• Keywords: Konvolucijska neuronska mreža; detekcija objekata; praćenje objekata; Tracktor metoda; otežani pomični prosjek; alpha - beta filtar; alpha - beta - gamma filtar; Kalmanov filtar
• Keywords (english): Convolutional neural network; object detection; object tracking; Tracktor method; weighted moving average; alpha - beta filter; alpha - beta - gamma filter; Kalman filter
• Language: croatian
• URN:NBN: urn:nbn:hr:168:392263
• 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 13:56:00