3D Corresponding Control Points Estimation using Mean Shift Iteration

Mean shift algorithm is widely used in 2D images. In this paper a novel 3D corresponding control points estimation using mean shift algorithm is proposed. This algorithm is not a simple extension from 2D to 3D, but computes the probability density function in each slice of the search region and connects them into a whole density function smoothed by Gaussian function. And then we calculate and compare Bhattacharyya coefficients to determine a new location of the trace point. A cylinder instead of ellipsoid is utilized as the search region to improve tracking accuracy. Also three revising methods different from the direct round-off way are proposed to modify the floating trace point. Experiment demonstrates the feasibility of this 3D mean shift algorithm and the effectiveness of the three revising methods.