3D model registration using CSIFT and POSIT

This paper describes a novel 3D model registration framework based on Canny edges, CSIFT key points, and Posit algorithm. First, we extract the Canny edges of the 3D model frame and camera frame, further handled by the Fast Fourier Transform, eliminating the impact of the pose differences. Then we use the particle filter to get the optimal position. Then we extract the CSIFT points using color invariant image instead of the gray image, retaining the color invariance and geometric invariance. K-d tree and Ransac algorithms are used to match the key points, which are used by POSIT algorithm to adjust model pose. Finally, we use a cup with different appearance to test the algorithm. The evaluation results support the potential of the proposed approach.