SOFTWARE IMPLEMENTATION USING TRANSFORMER WITH OPTICAL FLOW AND GEONET FOR IDENTIFYING PARAMETERS OF DYNAMIC OBJECTS

Today, interdisciplinary research in computer science and engineering has become increasingly relevant due to the growing demand for real-time data processing in object detection and tracking applications. The identification of dynamic object parameters plays a crucial role in various domains such as autonomous transportation systems, robotics, and surveillance. Effective automated acquisition and processing of video data represent a promising field for scientists and practitioners working in these interconnected disciplines. This research aims to enhance object detection and tracking processes by developing and implementing an information technology solution based on modern machine learning methods, including DETR (Detection Transformer), Optical Flow, and GeoNet. The research methodology involves designing software using Python programming language and modern libraries and frameworks for image and video processing. The DETR method was employed for precise object detection within video frames, Optical Flow was used to determine the direction and velocity of object movement, and GeoNet provided depth and geometric scene analysis. The proposed technology was tested on diverse video recordings depicting complex scenarios with dynamic conditions, such as varying lighting, object occlusions, and rapid motion changes. The results demonstrate the high accuracy and reliability of the proposed approach for identifying dynamic object parameters under various conditions. The integration of these methods significantly improved the precision and robustness of the detection and tracking system, particularly in challenging environments or low-quality video scenarios. The study concludes that the proposed information technology is effective and can be applied in practical fields such as autonomous systems, robotics, and video surveillance.