Abstract: Quick and accurate understanding of the disaster scene situation is the top priority in the disaster relief process. Usually, drones are used for site surveys when disasters occur, but it is difficult to associate the drone video with the actual geographic scene. To this end, this paper proposes a fusion method of mobile video and geographic scene. This method first uses ASift algorithm with affine invariance to detect feature points, and uses the RANSAC algorithm to iteratively remove noise points to calculate video and geography, optimal scene transformation matrix model parameters. Then the calculated perspective transformation parameters are then applied to the video data to restore the coordinates of the corners of the video. Finally, the corner coordinates of all video frames are obtained by interpolation to achieve an accurate fusion of video and DOM. The experimental results show that the shorter the interval frame that matches the video data, the higher the overall fusion accuracy, the standard deviation of the video and geographic scene fusion by this method is less than 10 m.

Key words: ASIFT algorithm, random sampling consistency algorithm, image matching, perspective transformation, image fusion