一种近实时的无人机视频流拼接方法  

A Nearly Real-Time UAV Video Frame Mosaic Method

作　　者：郑晖[1,2] 姜城[2] 孙敏[2] 黎晓东[2] 任翔[2] 刘磊[2] 

机构地区：[1]中国矿业大学(北京)地球科学与测绘工程学院,北京100083 [2]北京大学遥感与地理信息系统研究所,北京100871

出　　处：《地理与地理信息科学》2015年第5期1-5,共5页Geography and Geo-Information Science

基　　金：国家科技支撑计划项目(2012BAH27B02;2012BAK12B02)

摘　　要：为了解决无人机视频流的快速拼接问题,提出一种近实时的视频流拼接方法。该方法校正了倾斜拍摄带来的变形,改进了ORB算法特征提取和构建描述子的策略,提高了特征点的稳定性和描述子的判别性,在保证较高匹配率和足够匹配点对的同时提高了帧间变换模型的精度;最后为降低拼接误差,根据GPS数据和给定重叠度构建平衡二叉树,选取离二叉树节点最近的图像作为参考投影面完成拼接。实验结果表明:该方法减少了级联误差累积带来的影响,提高了拼接影像的质量,可用于无人机视频流的近实时拼接。In order to solve the problem of quick UAV video frame mosaic,a nearly real-time method is proposed.In this method,the image distortion resulted from oblique shooting is corrected,the strategy of feature extraction and descriptor construction in the ORB algorithm are improved,and the stability of feature points is increased,as well as distinction of descriptors.As a result,the precision of transformation models between frames is improved without decreasing the matching ratio or the number of matching points.In addition,the frame nearest to the node of the balanced binary tree is chosen as the projective plane,so that mosaic errors can be minimized.The experiment result shows that after applying this method to the mosaic processing,the cascading error is reduced,and the quality of result mosaic images is improved.What′s more,this method is qualified for nearly real-time tasks.

关 键 词：无人机 视频流拼接 倾斜拍摄 几何校正 级联误差 

分 类 号：P231[天文地球—摄影测量与遥感]