Back to search

Institutional Repository of Vilnius University

Neuroninio briaunų aptikimo taikymas lazerinei vektorinei projekcijai: vaizdo apdorojimo grandinės realizacija ir įvertinimas /

Abstract

dc:description

This thesis investigates the design and implementation of a laser controller that drives a physical ILDA-compatible laser projector from video content using neural network-based visual processing pipelines. The system accepts video input, extracts geometric line data through one of several processing pipelines, and transmits the result to a Helios DAC digital-to-analogue converter, which drives the projector at up to 30,000 points per second via the ILDA standard. Two distinct processing pipelines are designed, implemented, and evaluated: NBED and HED(Holistically-Nested Edge Detection). Both models are compared against Canny – a classic edge detection algorithm. Each pipeline at its beginning also uses SAM2(Segment Anything Model 2) for the purpose of separating trackable objects from the video background. The system is implemented as a modular four-stage pipeline: (0) CLAHE contrast enhancement and bilateral filtering for preprocessing, (1) edge detection (Canny / HED / NBED), (2) edge thinning (NMS / Zhang-Suen / Steger), and (3) polyline vectorization (Depth-First / Greedy Eulerian / Iarussi). Integration with SAM2 segmentation provides foreground masks, and a Helios DAC drives the physical ILDA-compatible laser projector. Empirical results show that NMS thinning achieves the lowest BR (HED+NMS+dp: BR = 0.075), while Steger ridge-linking generates excessive polyline fragmentation that saturates the 1000-point budget (BR approaches 1.0). Two predicted hypotheses were rejected by Wilcoxon paired signed-rank testing: NBED achieved statistically significantly lower wIoU than HED (p < 0.001, n = 28), and Steger produced statistically significantly higher BR than NMS (p < 0.001, n = 16). However, these results are not conclusive - the chosen metrics partially reward output sparsity rather than visual fidelity. No single pipeline can be recommended on the quantitative metrics alone.

Degree

thesis:*
Grantor dc:publisher
Institutional Repository of Vilnius University
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Šaulys, Mindaugas Gerardas,

Rights

dc:rights
Statement dc:rights
  • info:eu-repo/semantics/openAccess
Language dc:language
lit

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
Vilnius University
Base URL
epublications.vu.lt/oai
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Šaulys, Mindaugas Gerardas,. Neuroninio briaunų aptikimo taikymas lazerinei vektorinei projekcijai: vaizdo apdorojimo grandinės realizacija ir įvertinimas /. Institutional Repository of Vilnius University, 2026. https://repository.vu.lt/VU:ELABAETD308338102&prefLang=en_US