Three-dimensional Object Recognition Based On 3D Invariant

The three-dimensional object recognition is an important topic in computer vision research. It has huge economic and application value in the military and social domains. Since the half century, it has received the various countries' vision researcher's widespread attention, and the massive theories and the methods have proposed. But in conventional recognition methods, when the distance between camera and goal can't be approximate for being far limitlessly to the size of the goal, or when some sides and levels of formation of image do not keep the parallel relation, three-dimensional body's 2D image formation has the possibility to have the shape (structure) on distortion. In this case, the very big uncertainty will be produced, thus three-dimensional body accurate recognition is influenced seriously.In order to be able to recognize correctly and describe effectively the goal or scene in the formation of image, a method for the recognition of three-dimensional object is presented which is based on 3D invariance. A number of the characteristics and parameters, which are influenced by the image formation and the camera interior parameters, the projective invariant is adopted as the representative of the 3D invariance, are excavated from the identical object in the image group. These characteristic's extraction and the computation do not need to adjust and the reconstruct, only carry on directly on the 2D image. Firstly, the point-line characteristic pairs thoughout the point-line characteristic obtaining and the exact matching in the images. Then, on this basis, the 2D projective transformation as the tool which is established the view relation, through the virtual elements which are rationally presented, and many kinds of 3D invariant of complicated spatial structures can be neatly extracted by coming real elements and virtual elements. Finally, through the union of the 3D invariant model database and the hash to implementation of three-dimensional object recognitionestablish database and come to recognize the three-dimensional object. The results indicate that the invariant can effectively overcome influence of imaging condition, and the method could satisfy well three-dimensional object recognition demand.