Page 42 - Dynamic Vision for Perception and Control of Motion
P. 42

26     2  Basic Relations: Image Sequences – “the World”



                                     x     x O        angle for cameras); the rota-
                                                      tion takes place around an in-
                                            T         termediate y-axis, called node
                                                      axis k y . This already yields the
                    x o              P                new x-direction  x O, around
                          -z 0            \
                                                      which the final rotation takes
                   y  o                               place: The roll– or bank angle
                                                      I  indicates the angle around
                             I         I              this axis between the plane of
                                 T                    symmetry of the  vehicle and
                                                      the vertical plane. All of these
                z           y    z O  y O             angles are shown twice in the
                                                      figure for easier identification
             Figure 2.3. Transformation of a coordinate system  of the individual axis of rota-
                                                      tion.


            2.1.1.3 Scaling
            Due to Equation 2.1 scaling can be achieved simply by setting the last element in
            the HTM [lower right element p (4, 4)] different from 1. All components are then
            interpreted as scaled by the same factor p. This scaling is conveniently exploited by
            application to perspective mapping.

            2.1.1.4 Perspective Mapping
            Figure 2.4 shows some properties of perspective projection by a pinhole model. All
            points on a ray through the projection center P p are mapped into a single point in
            the image plane at a distance f (the focal length) behind the plane x p = 0. For ex-
            ample, the points Q 1, Q 2, and Q 3 are all mapped into the single point Q i. This is to
            say that the 3-D depth to the point in the real world mapped is lost in the image.
            This is the major challenge for monocular vision. Therefore, the rectangle in the
            image plane Re i may correspond both to the two rectangles Re 1 and Re 2 and to the
            trapezoids Trap 1 and Trap 2 at
            different ranges and with dif-
            ferent orientations in the real
                                                         x 1       x 2
            world. Any four-sided poly-  Image             Image
                                         plane   f   f
            gon in space (also nonplanar        Q i        plane          x
                                             Re i          (mirrored)  x 1  x 2  p
            ones)  with the corner  points           Re i
                                                           Re 1
                                         -y i
            on the four rays through the             y i   O
            corners given will show up as   y p  -z i   z i  P 1  O
                                          y 2            Q 1  P 2
            the same (planar, rectangular)          z p              O
                                                                    P 3
            shape in the image.            P =0         Trap 1  Q 2  Re 2  Trap 2
                                             p
              To get rid  of the sign      projection
            changes in the image plane     center                   Q 3 (x, y, z)
                                                                     O
            incurred by the  projection
            center P p (pinhole), the posi-  Figure 2.4. Perspective projection by a pinhole
            tion of this plane is mirrored   camera model
   37   38   39   40   41   42   43   44   45   46   47