无人机视频流倾斜影像快速拼接方法

为了解决无人机视频流倾斜拍摄快速拼接误差较大的问题,提出一种局部最优的视频流拼接方法。该方法首先利用透视投影的方法将倾斜影像纠正为近似正射影像,消除大倾角影像的几何变形问题;然后利用CUDA技术加速ORB算法,快速得到帧间变换模型;在拼接时动态选取基准面,利用Levenberg-Marquardt算法优化各影像变换到基准面的绝对单应性矩阵完成视频流的序列拼接。实验结果表明:利用该方法获得的拼接影像减小了倾斜拍摄带来的误差,提高了拼接影像的质量。

Oblique video flow mosaic based on unmanned aerial vehicle

In order to solve the problem of stitching error with UAV oblique photography video, a local optimum method is proposed. First, we recover the oblique image to rectified image through perspective transformation, which eliminates the geometric distortion; Secondly, use the CUDA technology to speed up ORB algorithm to get the transformational model quickly; Thirdly, dynamic selecting reference plane when stitching and optimizing the absolute homography matrix which transform images to reference plane with Levenberg-Marquardt algorithm. The experiment result shows that after applying this method to the mosaic processing, the error bring over from oblique photography is reduced, and the quality of result mosaic image is improved.