结合球心投影与线特征的点云与全景影像配准

发布时间：2018-07-14 13:08

【摘要】：当前国内外针对车载移动测量系统采集的点云数据和影像数据的配准优化问题多从硬件角度着手展开。即在配准优化前严格标定相机镜头,获得精确的相机内方位元素,在系统静止状态下使用激光扫描仪扫描标靶,使用相机拍摄标靶,利用摄影中心、标靶像点、标靶点云这三者共线的原理以及系统各部件间空间转换关系迭代优化配准参数。这些配准方法对外界目标合作标定物依赖性强、对相机内方位元素精度要求高,配准计算繁琐。本文提出了一种利用车载移动测量系统自身所采集的点云数据线特征和经形变纠正后全景影像数据线特征进行配准优化的方法。该方法不依赖外部检校设施,也不依赖相机内方位元素。首先对点云数据进行预处理。即使用与各站点云数据对应的GPS坐标参数,把点云由大地坐标系坐标转换为站心坐标系坐标,这一过程本文称作站心化。随后使用与各站点云数据对应的惯性导航系统参数,调整站心坐标系下点云的姿态,这一过程本文称为姿态矫正。在对点云数据进行预处理后,对全景影像指定区域形变进行纠正。本文采用球心投影纠正全景影像目标区域形变。球心投影等效于从球心发出的光束透过球面照射在与球面相切的平面上所成的像。除了具有方位投影的一般性质外,该投影的最大特性是保持直线特征的几何属性,即三维空间中的直线经球心投影后仍然是直线。经过球心投影处理后,全景影像指定区域形变即得到纠正。在形变纠正完成后,提取影像当中的线特征,并拟合线特征直线方程参数。提取点云当中的直线并投影到全景影影像,使数据维度降低。这样就可以使用像素思想处理点云。再使用球心投影将全景影像上的二维点云进一步投影到对应的全景影像子块,获得点云直线在该子块上的投影坐标初值。使用本文提出的直线方程与点云投影坐标初值匹配算法,将两者进行匹配。最后以线特征方程为约束条件,点云投影坐标初值为观测值,利用条件平差计算配准误差改正量,将配准误差改正量加入点云投影坐标初值,从而实现车载激光点云与全景影像的配准优化。
[Abstract]:At present, the registration optimization of point cloud data and image data collected by vehicle mobile measurement system at home and abroad is mostly carried out from the point of view of hardware. That is, the camera lens is strictly calibrated before registration and optimization, and the precise interior orientation elements of the camera are obtained. In the static state of the system, the laser scanner is used to scan the target, the camera is used to shoot the target, the photography center and the target image point are used. The collinear principle of target point cloud and the iterative optimization of registration parameters among the components of the system. These registration methods are highly dependent on the external object calibration and require high precision of the camera's interior azimuth elements, and the registration calculation is tedious. In this paper, a method of registration optimization based on point cloud data line features collected by vehicle mobile measurement system and panoramic image data line features corrected by deformation is proposed. The method does not depend on external calibration facilities, nor does it depend on the camera's internal azimuth elements. Firstly, the point cloud data is preprocessed. Even using the GPS coordinate parameters corresponding to the cloud data of each station, the point cloud is transformed from the geodetic coordinate system to the geodetic coordinate system. This process is called geodetic centroidization. Then the inertial navigation system parameters corresponding to the cloud data of each station are used to adjust the attitude of the point cloud in the center coordinate system. This process is called attitude correction in this paper. After preprocessing the point cloud data, the deformation of the designated area of panoramic image is corrected. In this paper, the spherical center projection is used to correct the deformation of the target area of panoramic image. The center projection is equivalent to the image of a beam of light emitted from the center of the sphere shining through the sphere on a plane tangent to the sphere. In addition to the general properties of azimuth projection, the greatest feature of the projection is to maintain the geometric properties of the linear feature, that is, the straight line in three-dimensional space is still a straight line after it is projected by the center of the sphere. After the spherical center projection, the deformation of the designated area of the panoramic image is corrected. After the deformation correction is completed, the line features in the image are extracted and the linear equation parameters are fitted. The line in the point cloud is extracted and projected to the panoramic image, which reduces the dimension of the data. In this way, you can use the pixel idea to deal with the point cloud. Then the 2-D point cloud on the panoramic image is projected to the corresponding panoramic image block by spherical projection, and the initial values of the projection coordinates of the point cloud line on the sub-block are obtained. Using the initial value matching algorithm of the linear equation and the point cloud projection coordinates, the two methods are used to match the linear equation and the point cloud projection coordinate. Finally, the line characteristic equation is taken as the constraint condition, the initial value of the point cloud projection coordinate is taken as the observation value, the registration error correction is calculated by using conditional adjustment, and the registration error correction is added to the initial value of the point cloud projection coordinate. In order to achieve vehicle laser point cloud and panoramic image registration optimization.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P235

【相似文献】