Point Cloud Depth Map and Optical Image Registration Based on Improved RIFT Algorithm

Abstract

In view of the unsatisfactory effect of the RIFT algorithm on local image registration, this paper introduces an improved RIFT algorithm based on the thin plate spline model for point cloud depth map and optical image registration method. To solve the problem of RIFT algorithm registration model, the thin-plate spline model is used instead of the rigid registration model to improve the algorithm. After image feature matching, the thin-plate spline is used to construct the image transformation model, and the image space transformation is decomposed into global affine transformation and local non-affine transformation, and the whole image and local mapping transformation are realized at the same time without local distortion. Experiments show that the improved algorithm can increase the CMR by 5%. The specific registration strategy is as follows: firstly, two kinds of data are preprocessed, and the image of the cloud depth map of the production point of the regular-grid resampling model is used. Then, the improved RIFT algorithm is used to extract corner points and edge points as registration elements, and Euclidean distance is used as similarity measure to achieve the registration of point cloud depth map and optical image, and then indirectly achieve the registration of laser point cloud and optical image. Finally, the registration accuracy is analyzed from the visual level and pixel level. The results show that the improved RIFT algorithm has favorable registration effect on point cloud depth map and optical image, and the proposed method has exceptional validity and reliability.