Globoko učenje za zaznavanje prstov v globinski sliki

MEDVEŠEK, LUKA (author), Čehovin Zajc, Luka (mentor)

Diplomska naloga poskuša izboljšati dosedanje postopke in metode zaznavanja gest in dotikov na večdotični površini. Starejše metode so za detekcijo uporabljale zaznavanje na podlagi oblike, barve ali rokavic s senzorji. Zaznavanje se ob zmogljivosti današnjih računalnikov lahko naredi s pomočjo konvolucijskih nevronskih mrež. Delo prikaže izdelavo zbirke podatkov in učenje nevronske mreže na globinskih slikah. Iz kvalitativnih ocen lahko vidimo, da je takšen pristop dovolj hiter in bi lahko bil dovolj natančen za takšno uporabo.

• Language: Slovenian
• Keywords: računalniški vid, večdotična površina, nevronska mreža
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2025
• PID: 20.500.12556/RUL-168004
• COBISS.SI-ID: 232796163
• Publication date in RUL: 24.03.2025

Secondary language

• Language: English
• Title: Deep learning for detecting fingers in depth images

Abstract

This project will try to improve the past procedures and methods of detecting touches on a multi-touch surface. Older methods used for detection gloves with integrated sensors, recognition based on shape or on color. Detection could be done with convolutional neural networks. This project presents the process of building data collection and learning of the neural network. From the measurements, we can see that this approach is fast enough to be processed in real-time and accurate enough to be used for the task.

• Keywords: computer vision, multitouch surface, neural network