A Multi-Image Mosaic Method for Farmland UAV Aerial Images via Reference Image Optimization

Abstract

In agricultural remote sensing, multi-image mosaic, which stitches the images captured by the camera and get an image with large range, is an efficient way to monitor and inspect a large range of farmland. However, one of the challenges is to accurately locate every component image into right positions in the resultant image. This locating process performs many times in multi-image mosaic. Apart from the first image, before locating any component image, an image that has already been located should be chosen as a reference image. A novel method named Reference Image Optimization (RIO) is proposed in this paper to select reference image in multi-image mosaic to reduce mosaic errors, where SIFT and RANSAC algorithms are applied to obtain the matched feature point pairs for every two-image mosaics, and the image with the largest number of matched feature point pairs at a particular range of direction relative to the locating image is chosen as the optimal reference image. In this paper, farmland is used as an application scenario to validate the method. The trait of farmland UAV aerial images is that there is a high similarity among different images, so mosaic error often appears when using traditional methods. The proposed method automatically stitches hundreds of high-resolution farmland UAV aerial images into a complete and seamless large image without resolution loss. The proposed method can be used to improve the accuracy of image mosaic technology of agricultural remote sensing.