The Application Of Projective Invariants Of Plane And Space In Object Recognition

Object recognition is one of the important parts of computer vision, and has been widely applied in the fields such as military affairs and automatization. The projective transform occurs in the 2-D image process, therefore in order to recognize object, we have to find projective-invariant shape descriptor of object. Description and recognition of object has been processed depend on the geometry model of plane target anciently. But people attaches more and more importance to the investigation of the perspective invariants with the development of the computer vision application, and becoming the one of the hot and difficult problems in the computer vision. Based on this study, this paper studies the recognition of two-dimensional and three-dimensional.Firstly, this paper reviewed current research about the invariant, and brief introduced the basic theory of the projective invariant in the object recognition. The research focus on the application that characteristic ratio and characteristic number in the match of characteristic points and the promotion of geometric invariants of plane curves in space. Main content:(1) a new invariant-characteristic ratio and characteristic number is used.Comparing the characteristic ratio and cross radio with the interference of noise,finding the characteristic ratio has smaller errors. The error of the characteristic ratio has smaller rate of increase than the cross ratio in the case of increasing the noise disturbance. This shows that the characteristic ratio has strongest ability of anti-noise. There will be the case that the points do not match but they have the same cross ratio when calculated the invariants using the cross ratio. In order to avoid this situation,we introduce the characteristic number to further determine. Characteristic number is a intrinsic nature of the curve,and it is projective invariant. We prove that the characteristic number can correct the miscarriage of justice to a certain extent by numerical experiments.It is an effective complement to the image invariant features. (2)Firstly, introduced the plane invariants that we used usually, then, obtain the intersection points of spatial lines and planes of space by 2D projective transformation,and the intersection points can be expressed by the projective transformation matrix and space points and space lines,so we have established the contact between the space elements and the plane elements, so extending the plane target of projective invariant to the spatial target. It expands the using range of geometric features invariants,bring the convenience to the identification of space objects.