GPU-based Real-time Image Mosaicing for UAVs

Abstract

Facing the problem of large amount of UAV image data, multiple redundancies and uneven quality, this paper proposes a GPU-based UAV real-time image mosaicing technology. First, the video stream data captured by the UAV is decoded to obtain frame images. Then, for these large number of frames images, pre-processing such as cache frame extraction and geometric correction is performed to obtain high quality and low redundancy UAV images. Extracting image feature points and matching them by SIFT, SURF, etc., and screening matching points by RANSAC algorithm. Finally, based on the screened matching points, image fusion is performed to obtain the UAV mosaicing image. The overall process is parallel accelerated by GPU, which is more efficient and real-time. The mosaiced images can be tile-cut and stored in chunks to achieve fast sharing and efficient application of image data.