Institutional Repository of Vilnius University
Bepiločio orlaivio vaizdo medžiagos vaizdinio triukšmo mažinimo metodai /
Abstract
dc:descriptionThis bachelor's thesis investigates video denoising methods for unmanned aerial vehicle footage. The objective is to compare classical and modern algorithms while improving FastNLMColored and ViDeNN methods by implementing parallelization and audio track preservation. A systematic literature review covers method evolution from classical filtering to deep learning solutions. FastNLMColored was enhanced with 7-core CPU parallelization, achieving 4.1× speedup (110s to 27s). ViDeNN was augmented with batch processing and GPU/CPU support, achieving 9.5% speedup. Real DarwinFPV BabyApe Pro drone footage was tested under three conditions: cloudy, night, clear weather. Quality was assessed using no-reference metrics (NIQE, BRISQUE, VIIDEO). Results show: ViDeNN most effective in night conditions (29.7% NIQE improvement), FastNLM h=25-30 best for clear weather (30.0% improvement), DaVinci Resolve UltraNR fastest (6.33s). Memory usage: ViDeNN (~2.2 GB), DaVinci Resolve (~8.1 GB), FastNLM (~12 GB). Each method has optimal applications: DaVinci Resolve for real-time, FastNLM for highest quality, ViDeNN for night conditions and limited resources.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Mikėnas, Aurimas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD239031756&prefLang=en_US