A novel approach to computer vision is proposed in this research work. Rather than designing ever more complex algorithms to extract given objects from images, the design of simple targets for given fast and robust algorithms is promoted. The targets and the computer vision algorithms are modelled analytically. The mathematical nature of this approach leads to optimized target structures. Two families of targets are investigated, line targets and dot targets, both with their respective extraction algorithms. The model of a Hough transform-based algorithm is developed for the line target and the performance of the vision system to detect the location of the target in 2D space, subject to noise, is analyzed. As well, the model of a Moment-based algorithm is developed for the dot target and the performance of the vision system to detect the pose of the target in 3D space, subject to noise, is analyzed. Design rules for the given target structures are established. The usefulness of the target design is demonstrated by practical examples which determine the location/pose of the targets affixed to objects. Target design can lead to simpler, faster and more robust computer vision systems for use in non-ideal industrial environments. Participating in this research were Dr. Bernd J. Kurz (Professor) and Dr. Agus Harjoko (formerly Ph.D. student). Project Status: Started: June 1989 Completed: July 1996
Agus Harjoko, Design of Targets for Computer Vision Systems, Ph.D. thesis, Fac. of Computer Science, Univ. of New Brunswick, Fredericton, N.B., Canada, July 1996 Harjoko, Kurz: Design of an Optimal Target for 3D Position and Orientation Estimation, Int'l. Journal of Microcomputer Applications, ISCA, Anaheim, Calif., submitted January 1996 Harjoko, Kurz: Design of an Optimal Target for 3D Pose Estimation, Proc. Int'l. Conf. on Computer Applic. in Industry and Engineering CAIE'94, ISCA, San Diego, Calif., Dec 1994, pp.135-138 Harjoko, Kurz: Target Design - A Method for an Accurate Pose Determination, Proc. 10th Conf. on Artificial Intelligence, IEEE, San Antonio, USA, May 1994, IEEE Comp. Society Press, pp. 242-252 Kurz: Target Design - A Simplifying Approach for Computer Vision, invited contribution to: Festschrift, Prof. Dr.-Ing. Egon Dorrer, published in honour of his 60th anniversary by Univ. der Bundeswehr, Munich, Germany, Feb. 1994, pp. 157-166 Harjoko, Kurz: Pose Estimation with Optimized Targets, Proc. 5th UNB Artificial Intelligence Symposium, Fredericton, Aug 1993, pp. 73-84 Harjoko, Kurz: Design of Optimal Targets for Computer Vision, Modell., Simulation and Identification MSI'92, IASTED, UBC, Vancouver, Aug 1992, pp. 80-85
Back to: Kurz's Research Page | Kurz's Home Page
Last revised: 6 June 1997 by Bernd Kurz bjkurz@unb.ca