An Efficient Geometric Transformation-Based Approach for Multi-UAV Image Stitching

UAV-based imaging systems often require real-time image stitching capabilities for tasks such as mapping, surveillance, and environmental monitoring. Traditional approaches typically rely on homography-based transformations, which, although robust, are computationally demanding. These methods often require reestimating transformation parameters for each video frame due to continuous camera motion, significantly increasing computational load and making real-time performance challenging, especially in resource-constrained drone applications. This study proposes an optimized image stitching method based on explicit rotation, scaling, and translation transformations derived from predefined anchor points. Designed for scenarios that involve relatively stable and flat scenes with known anchor markers, this approach removes the reliance on complex matrix operations. Evaluation on experimental footage demonstrates that the method reduces computational complexity, achieving processing times approximately 36% of those required by conventional homography-based methods. Visual assessments and alignment accuracy metrics indicate that the proposed method maintains the same alignment quality, making it particularly suitable for real-time UAV operations and embedded vision applications where computational efficiency is critical.