Registration Of Remote Sensing Images And LiDAR Data Based On Dual Quaternions

The registration of the LiDAR point cloud data and the optical image data is the process of finding the best match between the two data through coordinate transformation.It is the key point and difficulty in the research of urban 3D modeling technology.In actual production,in order to meet the requirements of large-area LiDAR point cloud and image registration,the large-scale sequence images are optimally incorporated into the object-based coordinate system to improve the accuracy of registration and registration method is required.The essence of registration is to solve the coordinate transformation relations of different data sets collected under different conditions(such as time,angle,sensor types,etc.).In the traditional registration geometry model,coordinate transformation representation methods such as Euler angle method and quaternion method are usually used to decompose the coordinate transformation into rotation and translation parts,and the registration model has certain requirements on the initial values.The inaccuracy of the value will cause a strong correlation between the rotational parameter and the translation parameter.The coupling error will be introduced in the process of separately solving the rotation and translation parameters,resulting in a larger mean-squared error of the least-square estimate and an iterative process can not converge.In order to solve the problems existing in the traditional registration,the paper proposes a registration method for LiDAR point clouds and sequence images based on dual quaternion geometry algebra.The expression of spatial line based on dual quaternion and its geometric topological relationship are studied.Dual quaternion is used to describe integrated rotation and translation process of the characteristic line and the dependence of dual quaternion parameter on the integrated rotation translation is analyzed.The characteristic lines on the sequence image and the control lines extracted from the LiDAR system are represented by linear coordinates,and coplanar condition equation based on dual quaternion is established.Using the collinear conditions of the characteristic point,a point-line joint registration set model of dual quaternion is established.The experimental results show that the registration geometric model based on the dual quaternion LiDAR and optical image of the joint dotted line feature,due to the integrated solution of rotation and translation process,avoids the coupling error between the two and reduces the correlation between the two.Therefore,it obtains higher registration accuracy than traditional method.The results of the paper provide a new idea for solving the problem of image registration in areas with lots of linear features such as urban areas.