METHODS OF INTELLIGENT VIDEO STREAM ANALYSIS FOR AUTOMATED EXTRACTION OF STRUCTURED DATA FROM OSD INTERFACES

Abstract

The paper investigates intelligent methods for automated telemetry data extraction from On-Screen Display (OSD) video streams. This task is critical for control systems where direct digital access to data protocols is restricted. The author proposes a comprehensive approach based on the synergy of classical image processing algorithms and modern deep learning architectures. The article details key stages: noise filtering, adaptive binarization, localization of regions of interest (ROI), and character recognition. Particular attention is paid to the comparative analysis of various optical character recognition (OCR) methods under limited computing resources for real-time processing. It is established that convolutional neural networks combined with frame segmentation ensure high resistance to dynamic backgrounds. The results confirm the feasibility of automating structured data collection with accuracy sufficient for further mathematical processing and visualization in information monitoring systems.