In order to solve the dual-Kinect external parameters calibration problem in scanning large object with dual-Kinect, we use an improved random sample consensus (RANSAC) method to extract the key parameters of sphere and plane model of the point cloud. The relationship between sphere center coordinate and normal vector of plane in different Kinect coordinate systems is analyzed. A new calibration method of dual-Kinect spatial position based on sphere and plane model is proposed. The calibration of the rotation translation matrix with large change in view and position movement is realized. Compared with the traditional iterative closest point (ICP) algorithm and improved ICP algorithm to calibrate dual-Kinect external parameters, the proposed method is faster and more accurate. Finally, the feasibility of this calibration method is verified by the actual model registration results. The proposed method provides a quick and accurate calibration scheme for dual-Kinect external parameters calibration with large change in view and position movement.