Binocular vision sequence image feature correspondence and motion detection based on cross ratio invariant

In this paper, we propose an efficient and convenient feature match method based on cross ratio invariant that can be applied in binocular vision sequence image motion detection task. On the basis of the principle that the cross ratio value of four points on the same a straight line is invariant in the perspective transform, we design a kind of special reference bar with five feature points. The reference bar is composed of one straight rigid bar with five reflective markers, and the distance between the reflective markers is calibrated in advance. By making use of the reference bar as the feature target, we can fast and accurately carry out the correspondence feature point match and object motion detection task according to the known four points cross ratio value. The epipolar theorem and the known cross ratio value are applied to eliminate the false match. The simulation experiments and experiments for real images have been conducted by using the proposed method. The result proves that the method is efficient when it is used in binocular vision sequence image motion detection task.