Research on multichannel image fusion and stitching technology based on wide-area imaging

Wide-area high-resolution imaging has become a crucial means for intelligence reconnaissance, surveillance, and global situational awareness due to its excellent concealment, real-time dynamic performance, high resolution, and wide coverage. It enables large-scale general surveillance, target searching, suspicious target identification, recognition, and tracking. Existing wide-area imaging technologies are primarily implemented through single-camera scanning imaging and multi-camera stitching imaging. The former suffers from a lack of time synchronization, making it suitable only for observing static or quasi-static scenes, and involves lengthy post-processing stitching, limiting its practical application. In contrast, the latter acquires large-field high-resolution images with simultaneity; image stitching is less affected by lighting changes or target motion, and it possesses video image acquisition capabilities, making it easy to achieve a large field of view and widely used in practice. This paper focuses on the issues encountered in multi-channel wide-area imaging, such as misalignment, ghosting, virtual shadows, stitching seams, linear structure distortion, and uneven illumination in stitched images. These problems arise because the objects being photographed are not on the same plane, cameras have depth-of-field errors, camera viewpoints are not co-located, and detector light response is inconsistent. The paper conducts research on lossless fusion and stitching technology for multi-channel images. Based on the pinhole imaging model, radial and tangential distortions between the lens and the imaging plane are analyzed, and distortion calibration methods and procedures are provided using a planar calibration method. On this basis, an imaging geometry model-based method is adopted to obtain positional relationships for multi-image stitching to achieve image registration and stitching. Furthermore, solutions for real-time, low-error, and illuminance gradient optimization stitching of multi-channel images are presented through global gain compensation, optimal seam line detection, and multi-band fusion. This study provides technical support for wide-area high-resolution imaging detection and multi-target detection and tracking.