一种融合单目视觉SLAM与GPS的无人机视频目标定位方法

A Method for Target Localization in Video of UAV with Combination of Mono Visual SLAM and GPS

提出一种融合单目视觉SLAM与机载GPS信息的无人机视频目标定位方法。首先利用单目视觉SLAM实时估计相机相对位姿,融合机载GPS数据,得到WGS84坐标系下的相机位姿估计;其次利用融合后的相机位姿将已知GPS路标投影到像平面,实现GPS路标与无人机监控视频的叠加显示;最后通过关键帧间的极线搜索匹配确定目标像素点的匹配点,并将其反投影到三维空间得其对应的GPS坐标,实现无人机监控视频中的GPS位置查询。实验验证了该方法的有效性、精确性与实时性。

The method combined with mono visual slam and GPS data for target localization in video of an UAV is presented in the paper. Firstly, the relative pose of the camera is estimated real-time through mono-visual slam and combined with GPS data acquired from UAV to get the global pose of the camera in WGS84 coordinate. And then the known GPS landmarks are projected into pixel coordinate and displayed in image frame. Finally, for any points in the image frame, the matched points are searched through the associated keyframes considering epipolar constraints and triangulated to get the GPS positions associated with them. The experiments verify the effectiveness, accuracy and real-time capability of the method.