International Symposium on Ubiquitous VR 2007                                                                                                        1




                    Cross-ratio Based Natural View Object
                         Recognition for Mobile AR
                                          Hyejin Kim, Sang-Goog Lee and Woontack Woo


                                                                          check the vanishing point or intersection of row lines in order
                                                                          to know whether the view is on one-side or two-side wall. Next,
                         I. INTRODUCTION                                  we calculate cross-ratio by using intersections of straight lines.
                                                                             3) The Off-line steps for Save as DB: We acquire cross-ratio
I    N upcoming ubiquitous computing environment, the
   number of smart objects will increase and various services
will be hosted by these smart objects. Therefore, it is necessary
                                                                          values for selected object and save it with direction and name
                                                                          information into database.
to show the information and services of smart objects in order               4) The On-line steps for Matching: If the view is one-side
to aid users for an easy access. As a part of this purpose, the           wall, then we calculate cross-ratio values for each object just as
vision-based AR (Augmented Reality) has played an important               show the method in description step. However, if it is two-side
role in the visual communication between smart environment                wall, we set the basis as boundary of wall and calculate
and users.                                                                cross-ratio values between objects at each wall.
   In this paper, we propose a method to recognize smart
objects based on invariant cross-ratios for indoor mobile AR.
This method supports natural view recognition, longer distance
(2m~3m) recognition and improving running speed without
marker compared with fiducial marker based or local feature
based method.

      II. CROSS-RATIO BASED NATURAL VIEW OBJECT RECOGNITION

  A. The Overall Procedures
   Figure 1 shows the overall procedures of cross-ratio based
                                                                                     Figure 1 The overall procedures of the proposed method.
natural view object recognition. It consists of input images,
off-line steps, on-line steps and output information-id and other           B. Cross-ratio
information of the objects. Like input images, in home                       Cross-ratio means ratio of ratio of lengths which make up of
environment, we can observe many rectangular shaped objects               four sets of collinear points. And it is invariant under projective
such as TV, windows, audio and shelves. For this reason, we               geometry [1]. If four points-x1, x2, x3, x4- are given in order,
re-define and use rectangular shapes of objects as natural view.          the cross-ratio is defined as equation (1). Also, if one of the
   The function of other steps is like following.                         points has a zero entity, it lies at infinity and simply cancels the
   1) Detection: We extract features which are quadrangle                 terms containing the point. For instance, if the second point
shape similar to marker based method. At this time, we also can
                                                                          have a zero entry, then x 23 = x 24 = ∞ . Thus, it cancels each
take partial quadrangle shape in order to use three or four points
for cross-ratio calculation.                                              other and a result of cross-ratio is defined as equation (2).
   2) Description: All quadrangle objects consist of the top and                                                  x1 x3 x 2 x 4
bottom row line and the left and right column line. Therefore,                  Cross ( x1 , x 2 , x3 , x 4 ) =                                    (1)
we make use of these straight lines from objects. Then, we                                                        x1 x 4 x 2 x3

  This research is supported by the Ubiquitous Computing and Network
(UCN) Project, the Ministry of Information and Communication (MIC) 21st                                                 x1 x3
Century Frontier R&D Program in Korea.                                                Cross ( x1 , x 2 , x3 , x 4 ) =                              (2)
  Hyejin Kim is with the Gwangju Institute of Science and Technology,                                                   x1 x 4
Gwangju 500-712, S.Korea (corresponding author to provide phone:
+82-62-970-2279; fax: +92-62-970-2204; e-mail: hjinkim@gist.ac.kr).
  Sang-Goog Lee is with the Catholic University of Korea, Bucheon-si,                                   REFERENCES
Gyeonggi-do 420-743, S. Korea (e-mail: sg.lee@catholic.ac.kr).
  Woontack Woo is with the Gwangju Institute of Science and Technology,   [1]   R. Hartley and A. Zisserman, Multiple view geometry in computer vision.
Gwangju 500-712, S.Korea (e-mail: wwoo@gist.ac.kr).                             Cambridge University Press, 2003.