Multimodal image registration is a fundamental task in computer vision and information fusion, supporting applications such as low altitude perception, medical imaging, remote sensing, and intelligent transportation. Gaps across modalities in imaging mechanisms, spectral responses, and geometric representations make cross-modal registration difficult in practice. Deployments face radiometric discrepancies, viewpoint variations, non-rigid deformations from platform motion, and mismatched resolutions. Recent progress in deep learning, cross-modal representation learning, and generative modeling has shifted conventional matching based pipelines toward end-to-end frameworks emphasizing fusion oriented modeling and joint optimization across tasks. This paper reviews the background, challenges, and methods for multi-modal image registration in low altitude scenarios and synthesizes feature-level and pixel-level approaches. We summarize integration into downstream tasks such as object detection, semantic segmentation, and image fusion, and discuss limitations, and future directions toward accurate, transferable, and controllable registration in complex environments.
Li et al. (Fri,) studied this question.