Target The Same 3d Feature Extraction

Object recognition based on images has been used in many fields of defense and civil economic construction. Extracting those features that show object's intrinsic characteristics and have strong adaptabilities is the kernel of recognition research all along. The traditional invariant features in translation,rotation and scaling transformation can't satisfy the needs of advanced object recognition. So the research about 3D invariance prospered in recent years is becoming one of the active research fields in computer vision. 3D invariant features are the features that will not change with the variety of viewpoints or poses,so it can solve the problems of mis-interpretation in perspective deformation. Thus it has significant potentials and interests in object recognition.Based on projective geometry,the research works about 3D invariance's extraction and application have been done in this thesis as following:(1) The basic theories and concepts in projective geometry are systematically summarized. It includes:the camera models of perspective imaging,projective collineation,cross ratio,a simple compare about invariance (invariant) among some geometry transformations,fundamental matrix,epipolar and epipolar line in epipolar geometry,and so on. (2) The calculation methods for 2D projective transformation are extended from points to multi-element,which includes points,lines,points lines and so on,to get the relationship between two projective planes. 3D invariants based on planar points and lines are summarized. In two-view case,a method is presented for calculation 2D projective transformation and construction the virtual elements by combination of spatial points and lines. So more 3D invariants of object's spatial structures can be extracted from real elements and virtual elements. (3) The geometric characteristics of 2D symmetric objects' in perspective images are lucubrated. The cross ratio of four lines is applied to describe feature points on object silhouette. For bilateral symmetry objects,the techniques on symmetry-point detection and symmetry-axis extraction are presented based on harmonic conjugation relationship. For rotated symmetry objects,the techniques on rotated units description and rotated symmetry center extraction are presented based on center invariants of objects. Further more,by using 3D invariants and 2D projective transformation,an approach to recover shape from part symmetry objects is realized on some conditions. (4) The special geometric restrictions like parallelism and perpendicularity implicated in vanishing elements are deeply studied. In single-view case,the theory is presented to interpret relationship between two planar lines by applying 3D invariants of vanishing points. In two-view case,the method on extraction vanishing point of an arbitrary spatial line is described,and moreover,the theory is presented to interpret relationship between two elements in 3D space,such as two lines,two planes,line and plane,and so on.Our approaches developed are applied to both synthetic and real images,and satisfied results are obtained.