Open Collections

A low cost three-dimensional vision system using space-encoded spot projection

University of British Columbia

Vision support plays an important role in intelligent robotic systems. It is often necessary to first recognize the location and the shape of the work piece before the robot arm can be directed to perform various tasks on it. A low cost three-dimensional (3-D) vision system based upon the space - encoded spot projection technique has been developed to measure the shape of a smooth, featureless curved surface. The vision system projects a 32x32 array of spots onto the measurement surface and uses a binary space-encoding scheme to incorporate the column address of these projection spots in a series of five projection patterns. Binary thresholding is implemented to extract the spot features in the images of the projection patterns. The image centroid positions and the decoded column addresses of the spot features, as well as the transformation matrices of the camera and slide projector are used to compute the spatial coordinates of the projection spots. This thesis establishes a calibration procedure to derive the transformation matrix of the camera using the measured spatial and image coordinates of a set of calibration image feature points. A similar calibration procedure is also established for the derivation of the transformation matrix of the slide projector. The accuracy of these matrices are substantially improved due to the utilization of the random sample consensus (RANSAC) algorithm to eliminate the gross error data points in the sample populations and allow only the good samples to be used in the derivation of the final transformation matrices. A surface reflectance model and a image centroid separation models are developed to describe the image intensity and spacings of the projection spots as a function of the measurement surface orientation. Using these models, the limiting orientations of a measurement surface with respect to the optical configuration of the vision system were found. Three reconstruction algorithms for fitting a surface to the scattered amplitude samples obtained by the spot projection system are described in the thesis. The suitability of these algorithms for the measurement of aircraft-wing surfaces is also discussed. The shape measurement results on a cylindrical surface showed that the average measurement error of points on a smooth featureless surface is less than a tenth of an inch, and the maximum error is less than a quarter of an inch in a total field of 20x20 inches.

Text

2010-07-12

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/26346

Electrical and Computer Engineering

University of British Columbia

Graduate