A target recognition technique employing geometric invariants

摘要

In this paper, we propose an efficient target recognition technique in the complex scenes. For the viewpoint invariant matching, the well-known geometric invariants, the invariants of five coplanar points and the invariants of five coplanar lines, which are invariant under the projective transform, are employed. To alleviate the computational complexity, the circular ordered matching technique and compatibility test are proposed, utilizing the geometric properties. Moreover, stability of the invariants for five coplanar points to the input noise is analyzed, and a method to select the stable invariants is also suggested. The performance of the proposed algorithm is evaluated on both various synthetic and real images. The results of the experiments show that the proposed matching technique is both effective and stable to the input noise for the target recognition in complex images.