车载全景相机的影像匹配和光束法平差

Image Matching and Bundle Adjustment Using Vehicle-based Panoramic Camera

提出一种针对车载全景影像序列的自动匹配和光束法平差方法。首先在单景鱼眼相机上进行多特征联合的初匹配;然后将同名特征投影到全景相机坐标系,并利用GPS/IMU数据进行粗差剔除和精匹配;再建立全景坐标下的透视投影几何模型,对匹配获得的连接点进行GPS/IMU辅助的光束法区域网平差,根据连接点的距离设定初始权,并通过选权迭代法剔除粗差,获得最终平差结果。试验结果表明,本文所提出的多特征匹配和全景坐标系平差的方法能够很好地处理全景相机的连接点提取和平差问题。

A new matching and GPS/IMU supported bundle adjustment method for vehicle-based panoramic image series is presented.Firstly,multi-features are extracted and preliminarily matched in the separate fisheye images,and the correspondents are projected to panoramic coordinate,and then gross error detection and second matching are done with GPS/IMU and LiDAR data.Secondly,the modified perspective transformation geometric model is constructed under panoramic coordinate,and by the model and GPS/IMU data bundle adjustment is carried out with the original weights set by the distance between object points to panoramic camera center.After gross errors excluded by varying weight method the adjustment results are gotten.The tests results proved that the multi-feature matching method and adjustment method under panoramic coordinate can goodly deal with the tie-points extraction and bundling for non-plane camera imagery.