指導教授：謝銘原 學 生：柯俊毅 學 號： M9820212 G.L. Foresti; F.A. Pellegrino; IEEE Transactions on applications and Reviews, Volume 34, Issue 3, Aug. 2004 Page(s):325.

指導教授：謝銘原 學 生：柯俊毅 學 號： M9820212

G.L. Foresti; F.A. Pellegrino; IEEE Transactions on applications and Reviews, Volume 34,  Issue 3,  Aug. 2004 Page(s):325 - 333

PPT製作：　（ 100%）

OUTLINE Introduction System description Automatic camera regulation Object classification Picking point extraction Experimental results Conclusion

Introduction This paper describes a vision-based system

that is able to automatically recognize deformable objects, to estimate their pose, and to select suitable picking points.

4

Fig. 1. General scheme of the automatic handling system.

5

Fig. 2. Functional diagram of the vision system.

Automatic Camera Regulation The goal of the camera regulation module is

to control the internal parameters of the camera (e.g., focus and aperture) in an automatic way, to allow acquisition of high-quality images.

7

2

1 1

TN , , for , Th (2)N N

x y

S x y S x y

2 2, ( , )

22 , , 1

I x y S x y I Ix y

i I x y i I x yx y

8

Fig. 4. Example of focus regulation: (a) original image acquired with a wrongfocus (focus step = 37) , and (b) the same image acquired with the best focus parameter (focus step = 44).

Object Classification

SOMs are a special class of ANNS based on competitive learning.

10

Fig. 5. General architecture of the applied neural classifier. SOM1 acts asfeature extractor, while SOM2 is the real classifier.

Picking Point Extraction

The output of the hierarchical SOM is characterized by several nonconnected regions of interest ( i.e., fur regions).

12

Fig. 6. Typical classification result obtained with n = 5(size of the macro-pixel), N = 6 and K = 5 (number of problem classes). (a) Original image, (b) output of the classifier. (c) pixels recognized as fur, (d) the detected connected components, (e) skeleton function of the detected blobs, and (f)

branch points

13

Fig. 7. (a) Four valid picking points extracted from the original image(b) three valid picking points extracted from an image acquired in realoperating conditions.

14

Fig. 8. (a) the original image containing 15 furs placed in a random way on a flat surface. (b) the output image, where five classes have been found. (c) the original image with the detected fur regions. (d) three valid picking points extracted from the original image.

Experimental results A. Camera Calibration B. Recognition and Handling of Furs C. Performance Evaluation

A. Camera Calibration

The aim of the vision system is to give to the robot arm the world coordinates of a correct picking point.

17

B. Recognition and Handling of Furs

Fig. 9. Further examples of the picking point procedure. Images where (a)-(d) valid picking points, (e) wrong picking points, and (f) any picking point have been detected.

C. Performance Evaluation After several experiments in real working conditions on the

fur tanning plant, we tuned the parameters of the picking point extraction module as follows:

(1)Blob Dimension [pixel]=300 , (2) Skeleton Dimension [pixel]=20 , (3) Symmetrical Factor=0.5 , and (4) Branches Dimension [pixel]=14 .

Table A summarizes the results of several tests done in order to evaluate the ability of the vision system to successfully detect picking points.

Results of several experimental tests done in order to evaluateThe ability of the vision system to successfully detect pickingPoints on a hear containing a different number of furs

20

TABLE A

The performance of the system has been evaluated experimentally. It achieves a successful picking point detection in more than 96% of the cases while location errors are less than 4 mm. initial implementation of the system in a fur tanning industry has retained the success rates achieved by the subsystems while future research is concentrated in the refinement of the manipulation strategy and the operation speed up.

21

Conclusion

22

Thanks for your attention!Thanks for your attention!

REFERENCES-1 [1] R. J. Campbell and P. J. Flynn, “A survey of free-form object representation and recognition techniques,” Computer Vis. Image Understanding, vol. 81, no. 2, pp. 166–210, 2001. [2] W. E. L. Grimson, Object Recognition by Computer—The Role of Geometric Constraints. Cambridge, MA: MIT Press, 1990. [3] H. Burkhardt, “Invariants for the recognition of planar contours and gray-scale images,” presented at the Workshop on Invariants for Recognition, in Conjunction with 2nd Eur. Conf. on Computer Vision, S. Margherita Ligure, Italy, May 19–22, 1992. [4] G. Taubin and D. B. Cooper, “Object recognition based on moment invariants,” in Geometric Invariance in Computer Vision, J. L. Mundy and A. Zisserman, Eds. Cambridge, MA: The MIT Press, 1992, pp. 375–397. [5] L. J. Van Gool, M. H. Brill, E. B. Barrett, T. Moons, and E. Pauwels, “Semi-differential invariants for nonplanar curves,” in Geometric Invariance in Computer Vision, J. L. Mundy and A. Zisserman, Eds. Cambridge, MA: The MIT Press, 1992, pp. 293–309. [6] Y. Lamdan and H. J. Wolfson, “Transformation invariant indexing,” in Geometric Invariance in Computer Vision, J. L. Mundy and A. Zisserman, Eds. Cambridge, MA: The MIT Press, 1992, pp. 335–353. [7] R. M. Haralick, H. Joo, C. Lee, X. Zhuang, V. G. Vaidya, and M. B. Kim, “Pose estimation from corresponding point data,” IEEE Trans. Syst., Man, Cybern. C, vol. 19, pp. 1426–1446, Nov./Dec. 1989. [8] R. M. Haralick and H. Joo, “2D-3D pose estimation,” in Proc. IEEE Int. Conf. Pattern Recognition, Rome, Italy, Nov. 1988, pp. 385–391. [9] A. Ghanei, H. Soltanian-Zadeh, and J. P. Windham, “A 3D deformable surface model for segmentation of objects from volumetric data in medical images,” Comput. Biol. Med., vol. 28, no. 3, pp. 239–253, 1998. [10] B. C. Vemuri and R. Malladi, “Constructing intrinsic parameters with active models for invariant surface reconstruction,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 15, pp. 668–681, July 1993.

REFERENCES-2 [11] C. Nastar and N. Ayache, “Deformable 3D objects: Using modes and FFT for a quantitative analysis of non rigid motion,” presented at the IEEE Workshop on Object Representation for Computer Vision, New York, Dec. 5–7, 1994. [12] , “Fast segmentation, tracking, and analysis of deformable objects,” in Proc. 4th Int. Conf. Comput. Vision, Berlin, Germany, May 11–14, 1993, pp. 275–279. [13] F. Leymarie and M. D. Levine, “Tracking deformable objects in the plane using an active contour model,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 15, pp. 617–634, June 1993. [14] G. Szekely, A. Kelemen, C. Brechbuhler, and G. Gerig, “Segmentation of 2-D and 3-D objects from MRI volume data using constrained elastic deformations of flexible Fourier contour and surface models,” Medical Image Analysis, vol. 1, no. 1, pp. 19–34, 1997. [15] H. I. Bozma and J. S. Duncan, “Modular system for image analysis using a game-theoretic framework,” Image Vis. Comput., vol. 10, no. 6, pp. 431–443, 1992. [16] J.W. Snell, M. B. Merickel, J. M. Ortega, J. C. Goble, J. R. Brookeman, and N. F. Kassell, “Model-based boundary estimation of complex objects using hierarchical active surface templates,” Pattern Recognit., vol. 28, no. 10, pp. 1599–1609, 1995. [17] K. V. Mardia, W. Qian, D. Shah, and K. M. A. Desouza, “Deformable template recognition of multiple occluded objects,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 19, pp. 1035–1042, Sept. 1997. [18] D. P. Huttenlocher, J. J. Noh, and W. J. Rucklidge, “Tracking nonrigid objects in complex scenes,” in Proc. 4th Int. Conf. Comput. Vis., Berlin, Germany, May 11–14, 1993, pp. 93–101. [19] G. Agam and I. Dinstein, “Geometric separation of partially overlapping nonrigid objects applied to automatic chromosome classification,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 19, pp. 1212–1222, Nov. 1997. [20] J. Segen, “Model learning and recognition of nonrigid objects,” Proc. IEEE Conf. Comput. Vis. Pattern Recognit., pp. 597–602, June 4–8, 1989.

REFERENCES-3 [21] J.Y.Weng, N. Ahuja, and T. S. Huang, “Optimal motion and structure estimation,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 15, pp. 864–884, Sept. 1993. [22] L. V. Tsap, D. B. Goldgof, S. Sarkar, and W. C. Huang, “Efficient nonlinear finite element modeling of nonrigid objects via optimization of mesh models,” Comput. Vis. Image Understanding, vol. 69, no. 3, pp. 330–350, 1998. [23] R. J. Holt and A. N. Netravali, “Motion of nonrigid objects from multiframe comparison,” J. Vis. Commun. Image Represent., vol. 3, pp. 255–271, 1992. [24] Y. Sato, M. Moriyama, M. Hanayama, H. Naito, and S. Tamura, “Acquiring 3D models of nonrigid moving objects from time and viewpoint varying image sequences—A step toward left ventricle recovery,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 19, pp. 253–259, Mar. 1997. [25] G. D. Sullivan, A. D.Worrall, and J. M. Ferryman, “Visual object recognition using deformable models of vehicles,” in Proc. Workshop Context- Based Vision, Cambridge, MA, June 1995, pp. 75–86. [26] N. Mandal and S. Payandeh, “Control strategies for robotic contact tasks: An experimental study,” J. Robotic Syst., vol. 12, no. 1, pp. 67–92, 1995. [27] D. W. Meer and S. M. Rock, “Experiments in object impedance control for flexible objects,” in Proc. Int. Conf. Robotics Automat., 1994, pp. 1222–1227. [28] K. B. Shimoga and A. A. Goldenberg, “Soft robotics fingeprint part I: A comparison construction material,” Int. J. Robotics Res., vol. 15, no. 4, pp. 320–333, 1996. [29] A. Howard and G. Bekey, “Intelligent learning for deformable objects manipulation,” in Proc. IEEE Int. Symp. Comput. Intell. Robotics Automat., Nov. 1999, pp. 15–20. [30] S. Haykin, Neural Networks: A Comprehensive Foundation. Englewood Cliffs, NJ: Prentice-Hall, 1996. [31] T. Kohonen, “Self-organized formation of topologically correct feature maps,” Biol. Cybern., vol. 43, pp. 59–69, 1988. [32] T. Kohonen, Self Organizing Maps. New York: Springer-Verlag, 1995.

REFERENCES-4 [33] N. Pican, E. Trucco, M. Ross, D. M. Lane, Y. Petillot, and I. T. Ruiz, “Texture analysis for seabed classification: Co-occurence matrices vs sel-organizing maps,” in Proc. OCEANS, Nice, France, Sept. 1998, pp. 424–428. [34] D. Kottow and J. Ruiz-del-Solar, “Texture segmentation by bio-inspired use of self-organizing networks,” Proc. Neural Comput., pp. 779–785, 2000. [35] D. A. Brown, I. Craw, and J. Lewthwaite, “A SOM based approach to skin detection with application in realtime systems,” presented at the Proc. British Machine Vision Conf., 2001, pp. 491–500. [36] Y. Deng and B. S. Manjunath, “Unsupervised segmentation of colortexture regions in images and video,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 23, no. 8, pp. 800–810, 2001. [37] L. Tang, L. Tian, B. L. Steward, and J. F. Reid, “Texture-Based Weed Classification Using Gabor Wavelets and Neural Network for Real-Time Selective Herbicide Applications,” Rep., ASAE Paper no. 991 151, 1999. [38] V. Murino, G. L. Foresti, and C. S. Regazzoni, “Adaptive camera regulation for investigation of real scenes,” IEEE Trans. Ind. Electron., vol. 43, pp. 588–600, Oct. 1996. [39] P. R. Wolf, Elements of Photogrammetry. New York: McGraw-Hill, 1983. [40] O. Faugeras, Three Dimensional Computer Vision. Cambridge, MA: The MIT Press, 1993. [41] C. Quek, G. S. Ng, and R. W. Zhou, “A novel single-pass thinning algorithm and an effective set of performance criteria,” Pattern Recognit. Lett., vol. 16, pp. 1267–1275, 1995. [42] M. K. Hu, “Visual pattern recognition by moment invariant,” IEEE Trans. Inform. Theory, vol. IT-8, pp. 179–187, Aug. 1962.