Abstract: In order to get true and high resolution texture for terrestrial LiDAR point clouds, the photo should be taken to all directions. This will cause the large angle problem for the registration of point clouds and the texture image. Meanwhile, the error from control points and scanner also creates the low registration accuracy. A robust registration method is proposed to solve these problems. First, rotation element model of registration parameter is derived by the barycenter model of the space similarity transformation, normal rotation matrix and the antisymmetric matrix. The registration result is treated as initial value. Then, the high accuracy registration parameters are obtained by improved Danish iteration method with variable weights based on collinearity equation. Experiment shows that new method is robust and accurate for the registration of any angle image and point clouds.