3D motion estimation for multi-object relative motion parameter and the centre of rotation using stereo sequence images

This paper proposes a general rigid motion model for estimating the multi-object 3D relative motion parameters and the centers of rotation based on stereo sequences images. It is assumed that the correct establishing correspondence between the matching points has been accomplished. For arbitrary two objects of these motion objects, firstly we set up three space reference frames respectively on the two objects and the left camera, and their space relation is described according to the coordinates of the correspondence points that lie on the two objects. Then, we design the two objects' 3D motion model through analyzing the rigid motion objects' character under the three reference frames. Thirdly, using these points' 3D space coordinates acquired by high speed stereo photography, we can linearly figure out the instantaneous relative 3D motion parameter and center of rotation through long sequence images after eliminating the motion affection of the other motion object. For heightening the motion estimation method's robust, RANSAC algorithm is applied to delete the outliers. Finally, simulation experiments are conducted using synthesized stereo sequence images corrupted by noise and the experiment results prove the model's correction and the algorithm's validity.