带状区域无人机影像自检校光束法平差方法

A Method for Self-Calibration Bundle Adjustment of Strip Region Images Captured by a UAV

针对带状序列无人机影像自检校空中三角测量时存在相机参数误差累积过大的问题,提出一种分类自检校(CSC)方法。该方法首先根据影像的GPS位置信息建立KD树,并利用K-Means进行自分类;然后对每类影像分别进行自检校光束法平差,将自检校得到的多组相机参数进行加权平均;最后进行全局自检校光束法平差。多组实验表明,CSC方法与室内检校场检校参数的像点畸变不符值均方根误差为0.5像素,检查点点位均方根误差为10.1 cm,且较Smart3D,VisualSFM和COLMAP软件能更精确地表示数据的原始姿态。综上,CSC方法可为带状区域无人机影像自检校空中三角测量提供一种有效的方案,具有较强的实践应用价值。

To solve the technical problem of excessive accumulation of errors of camera parameters during aerial triangulation self-calibration of strip sequence images captured by a UAVa Classified Self-Calibration(CSC)method is proposed.Firstlybased on the GPS information of the imagesa KD-tree is set upand K-Means is used for automatic classification of the images.Thenthe self-calibration bundle adjustment of each type of images is conducted respectivelyand weighted average is made to several groups of camera parameters obtained through self-calibration.Finallythe global self-calibration bundle adjustment is conducted.Multiple sets of experiment show thatthe RMSE of the parameter difference between the CSC method and the calibration method conducted in an indoor calibration field after image point distortion removal is 0.5 pixelsand the RMSE of the checkpoint position is 10.1 cm.Compared with that of Smart3DVisualSFM and COLMAP softwarethe results of aerial triangulation of the proposed method could more accurately represent the original pose of the data.In conclusionthe CSC method provides an effective scheme for the aerial triangulation self-calibration of strip sequence images captured by a UAVwhich has high practical application value.