Multi-overlapped based global registration of UAV images

A new global registration method based on multi-image matching is proposed for image mosaic. In this study a manifold learning algorithm is first used to obtain multi-images matching results, on which a globally mosaic image coordinate system is then iteratively achieved and utilized to calculate global registration parameters of each image. Testing results indicate that the proposed method has a higher performance than traditional frame-to-frame registration methods in the error accumulation and registration accuracy of mosaic images.