Planar Object Recognition Based On Projective Invariants

Vision is the best perception system of human. Vision invariants for planar object recognition catch our attention. This thesis analyzed thoroughly the affine invariant features and the projective invariant features of planar objects and a novel method based on the projective invariants was deeply studied. Simulation experiments and performance analysis are both given.We introduced the extractions of vision invariants and the related conclusions, which are compared and contrasted. We also present a canonical frame construction for determining projective invariants for the non-convex smooth plane curves. This method requires no camera calibration or pose information, and models are generated and verified directly from images. The application of the method uses the object's boundary, so the edge detection and segmentation were studied. The features are stable and discriminable. The invariants are semi-local rather than global, which promotes tolerance to occlusion. The considerable redundancy is possible for the object whose boundary has complex structure, which enhances the reliability of object recognition.