Visual Object Tracking on a Mobile Platform

Tenjović, Dušan (author), Lukežič, Alan (mentor)

This thesis presents a real-time object segmentation and tracking system integrated into the ViCoS Cube demonstration cell. It implements a segmentationbased tracker DAM4SAM that uses SAM2.1 model combined with a distractoraware memory. The tracker provides precise pixel-level masks and object centroids for closed-loop pan-tilt-zoom (PTZ) camera control. A Docker-based, publish/subscribe architecture using Echolib enables modular integration of camera streams, tracking, and GUI. The system is evaluated in a controlled environment, analyzing accuracy, robustness, and speed with different hardware and camera configurations. Finally, the system’s ability to follow and segment a moving object is demonstrated in real-time, closing the loop between perception and camera actuation. This project brings cutting-edge segmentation and tracking methods to an interactive educational setup.

• Language: English
• Keywords: object tracking, image segmentation, SAM, DAM4SAM, computer vision, real-time processing, mobile platform, PTZ camera
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2026
• PID: 20.500.12556/RUL-179586
• COBISS.SI-ID: 270250243
• Publication date in RUL: 18.02.2026

Secondary language

• Language: Slovenian
• Title: Vizualno sledenje objektov na mobilni platformi

Abstract

V tej diplomski nalogi je predstavljen sistem za realnočasovno segmentacijo in sledenje objektov, integriran v demonstracijsko celico ViCoS Cube. Implementiran je segmentacijski sledilnik DAM4SAM, ki uporablja model SAM2.1 v kombinaciji s spominskim modulom, odpornim na motilce. Sledilnik zagotavlja natančne maske na ravni slikovnih točk in položaje centrov objektov, ki se uporabljajo za zaprtozančno krmiljenje kamere s pan-tilt-zoom (PTZ) mehanizmom. Modularna arhitektura, zasnovana na Dockerju in komunikaciji tipa publish/subscribe prek Echoliba, omogoča preprosto integracijo video tokov, sledenja in grafičnega vmesnika. Sistem je ovrednoten v kontroliranem okolju, pri čemer se analizira natančnost, robustnost in hitrost delovanja pri različnih konfiguracijah strojne opreme in kamer. Na koncu je prikazana tudi realnočasovna demonstracija sledenja in segmentacije premikajočega se objekta, kjer sistem zaključi zanko med zaznavanjem in aktuacije kamere. Projekt tako v interaktivno izobraževalno okolje prinaša sodobne pristope segmentacije in sledenja objektov.

• Keywords: sledenje objektov, segmentacija slik, SAM, DAM4SAM, raˇcunalniˇski vid, realnoˇcasovna obdelava, mobilna platforma, PTZ kamera