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

Abstract

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