The Registration Of Vehicle-based Point Clouds And Panoramic Images Based On Linear Features And Gnomonic Projection

Currently,the registration optimization of point clouds data and images data acquired with the vehicle mobile measurement system is mostly treated from the aspect of hardware at home and abroad,which means to strictly calibrate the camera lens before registration optimization to obtain precise orientation elements in the camera.When the system is at still state,the target is scanned with laser scanner and taken with the camera,and the registration parameters are optimized according to the collineation principle of photographic center,target image point and target point clouds and the spatial switching relation iteration of different components.These registration methods have strong dependency to the cooperative calibration of the external object,and high requirements on the precision of orientation elements in the camera with complicated registration calculation.This thesis proposes a kind of method for registration optimization with the point clouds data linear features acquired by the vehicle mobile measurement system and the panoramic images data linear features after deformation correction.This method does not depend on the external calibration and verification facilities and the orientation elements in the camera.First,the point clouds data are preprocessed.The point clouds are transferred from the coordinate of geodetic coordinate system into that of topocentric coordinate system with the GPS coordinate parameters which is corresponding with the point clouds data of different stations,this process is named as topocentric-orientation in this thesis.Then the point cloud pose under the topocentric coordinate system is adjusted with the inertial navigation system which is corresponding with the point clouds data of different stations,and this process is named as post correction in this thesis.After the preprocessing of point clouds data,the deformation of panoramic images in the designated region is corrected.This thesis adopts gnomonic projection to correct the deformation of the targeted region in the panoramic image.Gnomonic projection is equivalent to the image formed on the plane which is tangent with the spherical surface with the light beam transmitted from the spherical center through the spherical surface.Besides the common properties of azimuthal projection,the most important advantage of this project is to keep the geometric properties of linear features,which means that the straight line in the 3D space still keeps at straight line after gnomonic projection.After processing via gnomonic projection,the deformation in the designated region of panoramic image is corrected.After deformation correction,the linear features in the image are extracted and the fit linear features linear equation parameters.The straight lines in the point clouds are extracted and projected to the panoramic image to reduce the data dimension.Then the point clouds can be processed with pixel idea.Gnomonic projection is adopted to project the 2D point clouds of panoramic images to corresponding panoramic image subblock to get the initial value of projection coordinate of the point clouds straight line in this subblock.Both of them are matched with the algorithm of matching the linear equation proposed in this thesis and the point clouds initial value of projection coordinate.Finally,the linear features equation is used as the constraint condition,point clouds initial value of projection coordinate is used as the observed value,the condition adjustment is adjusted for calculating the correction of registration error,and the correction of registration error is added into the point clouds initial value of projection coordinate,thus achieving the registration optimization of vehicle laser point clouds and panoramic images.