Institutional Repository of Vilnius University
Išryškinimo bei triukšmo pašalinimo metodų panaudojimas, kad geriau veiktų objektų atpažinimas analogo tipo filmukuose /
Abstract
dc:descriptionThis thesis explores object recognition in analog-type video footage, which typically suffers from low resolution and high noise levels. To improve detection accuracy, various denoising and upscaling techniques—such as BM3D, RealPLKSR, and Real-ESRGAN—were applied. Experiments were conducted using the YOLOv8 object detection model, trained on differently pre-processed datasets. The results showed that preprocessing alone was insufficient—best performance was achieved when models were trained specifically on analog video data. Moreover, it was found that the unique noise characteristics of each camera significantly affect recognition accuracy, highlighting the need for device-specific training. The study concludes that effective use of AI models in analog footage requires both appropriate preprocessing and customized training.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Pupelis, Gvidonas,
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:ELABAETD239025044&prefLang=en_US