Image-Only Real-Time Incremental UAV Image Mosaic for Multi-Strip Flight

Fast and Efficient Method for Large-Scale Aerial Image Stitching

“…Moreover, the predefined overlapping ratio may vary significantly during UAV flights owing to wind. In a previous study, the overlapping ratio was set as follows: 75% [20], at least 75% frontal overlap (with respect to the flight direction) and at least 60% side overlap (between flying tracks) [35], or 80% frontal overlap and 60% side overlap [36]. Additionally, it is recommended that aerial images be captured with at least 80%-85% overlap to obtain the best output quality for forest or densely vegetated areas [35], [36].…”

“…If the input images are too dense or several images share the same overlapping region, the dense correspondence may lead to mis-registration [8]. Consequently, the stitching performance of densely overlapped input images becomes inadequate [20], [25]. To address this problem, we propose a fast adaptive selection algorithm to eliminate unnecessary input images that are densely overlapped with other images.…”

“…The problem of aerial image mosaic composition has received considerable attention in recent years [9], [18]- [25]; real-time stitching has been considered in several studies [9], [18]- [20]. Bag-of-words image representation was employed in [9] to determine the overlapping and rapidly extract correspondences among images.…”

“…However, perspective distortion caused by accumulated errors has been observed in these methods [9], [18], [19]. Another real-time incremental stitching method was proposed in [20], where the homography transformation of the current frame to the key frame was estimated based on local optimization. In [21], the aerial images were stitched in real-time using the incrementally expanded road map, and then, the output mosaic image was used for geolocalization.…”

“…Previous studies on image stitching have been focused on the simple case of stitching two images [10], [11] or multiple images located on the same row to generate panoramic images [12]- [17]. The problem of generating mosaic images from UAV images is that numerous input images taken over a wide area are usually required [9], [18]- [20], [24], [25].…”