Machine Learning-Based Visual Localization of a Moving Object, Trajectory Construction, and Development of Forecasting Hardware and Software

This paper explores the application of machine learning techniques for visual localization, trajectory prediction, and the development of FPGA-based forecasting hardware and software. Integrating computer vision with real-time data processing, the proposed system addresses challenges in dynamic environments, enhancing accuracy and efficiency. The proposed methodology leverages dual-modality drone detection using parallel RGB and infrared (IR) processing pipelines, providing a comprehensive framework that combines robust YOLO models, efficient FPGA implementations, and scalable architectures for real-world applications. By integrating two specialized YOLO models one for RGB and one for IR imagery - with a late fusion strategy, the system achieves high accuracy and performance even in challenging scenarios such as occluded or GPS-denied environments. Additionally, this research highlights the deployment of optimized forecasting software for hardware acceleration, enabling real-time processing at the edge.