Automatic machine vision calibration using statistical and neural network methods

Smith, Lyndon N.; Smith, Melvyn L.

Abstract

A methodology is presented for camera calibration that is designed to improve the accuracy of machine vision based object measurement systems. The regression and artificial neural network techniques studied are considered to be complimentary rather than competitive. Neural networks have been identified as being particularly useful for the precise modelling of non-linear response, and offer the additional benefits of being non-prescriptive and generally applicable to factors such as radial lens distortion, manufacturing errors and minor camera misalignments. The combination of these modelling techniques within automated program control strategy is suggested as a new approach for straightforward and accessible machine vision calibration. The method has particularly good application to vision metrology and reverse engineering tasks. A demonstrator system has been constructed, employing a scanning laser line and vision system for object measurement in three-dimensions. Experimental results are presented along with a demonstration of the reduction in measurement error that can be attained through the application of regression analysis and artificial neural network modelling. © 2005 Elsevier B.V. All rights reserved.

• Journal Article Type: Article
• Publication Date: Sep 20, 2005
• Journal: Image and Vision Computing
• Print ISSN: 0262-8856
• Publisher: Elsevier
• Peer Reviewed: Not Peer Reviewed
• Volume: 23
• Issue: 10
• Pages: 887-899
• DOI: https://doi.org/10.1016/j.imavis.2005.03.009
• Keywords: automatic calibration; radial lens distortion, linear regression, artificial neural network
• Public URL: https://uwe-repository.worktribe.com/output/1047629
• Publisher URL: http://dx.doi.org/10.1016/j.imavis.2005.03.009
• Additional Information: Additional Information : This work resulted from a more in-depth analysis of an image distortion problem in connection with research into calibration of a vision based metrology system developed by the author for Quantronix. The work went well beyond previous techniques since it does not assume that lens distortion is axis-symmetrical. It is relevant to metrology applications, but can also be applied in any field involving image capture, with a wide angle lens. The paper was listed among the journal's 25 most read articles. The metrology system is now being marketed and installed worldwide. (C Skeen, Quantronix Incorporated, Farmington Utah.)