Low contrast point cloud data registration based on SIFT feature

Aiming at registration of low contrast point cloud data, which is measured with multi view that changes largely and is accompanied with noise from complex product, a data registration method based on SIFT feature is proposed, which focus on robust feature point identification. Point cloud data x, which reflects the product shape and relationship, is divided into the low frequency part and the high frequency part. Spatial location change has little effect to the low frequency part, while the high frequency part changes greatly with spatial location and product geometric shape characteristics. Gauss homomorphic filter is designed to reduce the low frequency part and increase the high frequency part, improving the contrast of point cloud data. SIFT feature vector of point cloud data is extracted. SIFT feature vector matching is realized with Euclidean distance as the similarity measure to achieve matching points of point cloud data. Quaternion is used to estimate the parameters of point cloud data registration and calculate rotation matrix and translation matrix, achieving low contrast point cloud data registration. Registration example of point cloud data from a brake shell shows the validity of the proposed methods.