International Journal of Computer Vision

, Volume 2, Issue 4, pp 395–415 | Cite as

Motion from flow along contours: A note on robustness and ambiguous cases

  • Fredrik Bergholm


The projected contours of a rigid body in motion define a velocity field, a flow in the image plane, where the tangential components are invisible. We investigate robustness when estimating motion parameters from the normal components of velocity along contours.

Ambiguous curves are contours for which no unique motion can be recovered from normal flow. We show that they are field lines of some extra velocity field, which causes an invisible motion when added on top of the actual velocity field. (These field lines coincide with the contour.) This is an analytical description of what Waxman and Wohn (1984) called “the aperture problem in the large,” which actually was studied by Lie (1893) under the concept of “self-projective curves,” if we specialize to a planar surface patch. We make a catalogue of the class of ambiguous curves.

We study the robustness of motion parameters, obtained by least-squares estimation, by calculating the a mount by which noise (in the normal flow) is magnified. For simple curves and a motion model with four unknowns-two translation components, rotation and expansion-there are explicit formulas relating the robustness of motion parameters to how large a portion of the contour we use. Sometimes it helps to use more global information (a larger portion of, e.g., a closed contour), and sometimes there is no way out, because we are working with contour segments that too closely resemble ambiguous curves. Neither is there any hope to recover the true velocity field from normal flow locally, since any contour falls into one of the “ill-conditioned contours.”


Velocity Field Field Line Motion Parameter Normal Flow Surface Patch 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    G.Adiv, “Determining 3-D motion and structure from optical flow,” Coins Technical Report 84-07, Univ. of Massachusetts at Amherst, April 1984.Google Scholar
  2. 2.
    S.Carlsson, and C.Reillo, “Contour based representation of the displacement field for motion compensated image coding,” Proc. ICASSP Conf., Tokyo, April 1986.Google Scholar
  3. 3.
    J.F. Canny, “Finding edges and lines in images,” TR No. 720, Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, MA., 1983.Google Scholar
  4. 4.
    J.F. Canny, “A computational approach to edge detection,” IEEE Trans. PAMI 8 (November 1986).Google Scholar
  5. 5.
    L.S. Davis, Z. Wu, and H. Sun, “Contour-based motion estimation,” Comput. Vision Graphics and Image Processing. 23:303–326, 1983.Google Scholar
  6. 6.
    E. Hildreth, “Computations underlying the measurement of visual motion,” Artificial Intelligence 23: 309–354, 1984.Google Scholar
  7. 7.
    H.C. Holland, The Spiral After-Effect. Pergamon Press: Oxford, 1965.Google Scholar
  8. 8.
    B.K.P. Horn, and B.G. Schunck, “Determining optical flow,” Artificial Intelligence 17: 185–203, 1981.Google Scholar
  9. 9.
    K.Kanatani, “Structure from motion without correspondence: General principle,” Proc. DARPA Image Understanding Workshop, Miami Beach, FL, pp. 107–116, December 1985.Google Scholar
  10. 10.
    J. Kmenta, Elements of Econometries. Macmillan: New York, 1971.Google Scholar
  11. 11.
    R.Lenz, “Estimation of general 2-D motion parameters in TV-scenes,” Proc. 7th Intern. Conf. on Pattern Recognition, Montreal, vol. I, pp. 546–548, 1984.Google Scholar
  12. 12.
    Lie Sophus, Vorlesungen über Continuerliche Gruppen, Chelsey Publ. Company, 1893.Google Scholar
  13. 13.
    D. Marr, and E. Hildreth, “Theory of edge detection,” Proc. Roy. Soc. London B-207: 187–217, 1980.Google Scholar
  14. 14.
    C. Reillo, and S. Carlsson, “Contour segment motion estimation,” TRITA-TTT 8506, Royal Institute of Technology, Stockholm, Sweden, November 1985.Google Scholar
  15. 15.
    M. Subbarao, and A.M. Waxman, “On the uniqueness of image flow solutions for planar surfaces in motion,” Center for Automation Research, Tech. Rept. CS-TR-1485, University of Maryland, College Park, April 1985.Google Scholar
  16. 16.
    R.Y. Tsai, and T.S. Huang, “Estimating three-dimensional motion parameters of a rigid planar patch,” IEEE Trans. Acoustics, Speech, and Signal Processing 29(6): 1147–1152, 1981.Google Scholar
  17. 17.
    R.Y. Tsai, T.S. Huang, and W.L. Zhu, “Estimating three-dimensional motion parameters of a rigid planar patch, II: Singular value decomposition,” IEEE Trans. Acoustics, Speech, and Signal Processing 30(4): 525–534, 1982.Google Scholar
  18. 18.
    G.Simmons, Differential Equations with Applications and Historical Notes. TATA McGraw-Hill Publishing Company Ltd, 1972.Google Scholar
  19. 19.
    H. Wallach, A. Weisz, and P.A. Adams, “Circles and derived figures in rotation,” Am. J. Psychol. 69: 48–59, 1956.Google Scholar
  20. 20.
    A. Waxman, and S. Ullman, “Surface structure and 3-D motion from image flow: A kinematic analysis,” CAR-TR-24, CS-TR-1332, Center for Automation Research, Univ. of Maryland, Artifical Intelligence Lab., MIT Cambridge, MA, October 1983.Google Scholar
  21. 21.
    A. Waxman, and K. Wohn, “Contour evolution, neighbourhood deformation and global image flow: Planar surfaces in motion,” CAR-TR-58, CS-TR-1394, April 1984, Center for Automation Research, Univ. of Maryland; also in Intern. J. Robotics Res. 4:72–94, 1985.Google Scholar
  22. 22.
    K. Wohn, L.S. Davis, and P. Thrift, “Motion estimation based on multiple local constraints and nonlinear smoothing,” Pattern Recognition 16(6): 563–570, 1983.Google Scholar

Copyright information

© Kluwer Academic Publishers 1989

Authors and Affiliations

  • Fredrik Bergholm
    • 1
  1. 1.Royal Institute of Technology, NADAStockholmSweden

Personalised recommendations