Image Stitching Methods for Enhanced Multicamera Computer Vision

This paper provides a discussion of advancements in image stitching techniques and presents a hands-on  
implementation and testing of selected methodologies. Existing approaches are analyzed, ranging from classical homography based methods to modern deep learning techniques, followed by the development of three progressive prototypes: simple homography based stitching, parallax-tolerant seam-driven refinement, and hybrid classical-deep integration using LoFTR correspondence matching.

The research establishes foundations for transforming static prototypes into real-time streaming pipelines capable
of simultaneously processing N-input multicamera systems. These image stitching pipelines appear suitable for future enhancement of multicamera computer vision systems and AI model input processing.

DOI: 10.1109/ICETA67772.2025.11280234
‍