Automated CAD-Based Machine Vision

Faculty: Linda G. Shapiro and Robert M. Haralick
Current Students: Mauro Costa, Bharath Modayur, Kari Pulli, and
                  Adnan Mustafa
The goal of this project is to develop an automated vision system for inspection and robot guidance that bridges the gap from CAD models to machine vision algorithms. The PREMIO system converts a CAD model to a model that is appropriate for machine vision, uses the vision model to predict features that will appear in images under various lighting and other environmental conditions, and uses the predictions to guide a matching procedure that finds correspondences between image features and model features for estimation of position and orientation. The ICE system determines the best positions for sensor and light source for a given inspection task. The Automated Inspection System allows the user to select an object and its features to be inspected and an inspection task. It then performs the specified dimensional inspection on a given image of the selected object.

The Shape from Color Photometric Stereo System determines 3D shape of objects in a scene from two color images of the scene taken from the same viewpoint with two different lightings. The system is being used as part of a color object recognition procedure . The Parallel Object Recognition System matches point or line-segment models of 2D and 3D objects to features extracted from images, using a new matching procedure implemented as a parallel algorithm on both SIMD and MIMD machines as well as on UNIX workstations. A new system for active analysis of multi-object scenes is under development. The system uses a single, movable camera and several light sources. It generates hypotheses about the scene from several images taken from a single viewpoint with different light sources. The hypotheses will then be used to determine an action such as moving the camera or light sources and taking more images with which to verify or disprove the hypotheses. After several iterations of image acquisition, image processing, analysis, actions, the system will produce an explanation of the scene in terms of the models that g, analysis, actions, the system will produce an explanation of the scene in terms of the models that it finds present and their positions in the scene.

  • Yi, S., R. M. Haralick, and L. G. Shapiro, ``Optimal Sensor and Light Source Positioning for Machine Vision,'' to appear in CVGIP: Image Understanding, 1995.
  • Christensen, P. H. and L. G. Shapiro, ``Three-Dimensional Shape from Color Photometric Stereo,'' International Journal of Computer Vision}, 1994.
  • Yi, S., R. M. Haralick, and L. G. Shapiro, ``Error Propagation in Machine Vision,'' Machine Vision and Applications, Vol 7, 1994, pp. 93-114.
  • Modayur, B. R., L. G. Shapiro, and R. M. Haralick, ``Visual Inspection of Machined Parts,'' in Advances in Image Processing and Machine Vision, J. Sanz, ed., Springer Verlag, 1994.
  • Shapiro, L. G., ``View Class Representation and Matching of 3D Objects,'' in Visual Form: Analysis and Recognition, C. Arcelli, L. Cordella, and G. Sanniti di Baja, eds., New York: Plenum Press, 1992, pp. 479-494.
  • Camps, O. I., L. G. Shapiro, and R. M. Haralick, ``Image Prediction for Computer Vision,'' Three-dimensional Object Recognition Systems, A. Jain and P. Flynn (eds.), Elsevier Science Publishers BV, 1992.
  • Costa, M. S., R. M. Haralick, and L. G. Shapiro, ``Optimal Affine-Invariant Matching: Performance Characterization'', SPIE Symposium on Electronic Imaging, San Jose, CA, February, 1992, pp. 22-31.