Development of DSL Three-Dimensional Calibration Technique, 10-9535

Printer Friendly Version

Principal Investigators
Ernest A. Franke
Michael P. Rigney
Michael Magee
Ian Meinzen

Inclusive Dates:  04/06/05 – 08/06/05

Background - The DSL three-dimensional (3D) measurement technique, developed under an earlier IR&D project, includes a calibration procedure based on a pinhole lens model. This procedure was used for 2 years but occasionally produced a calibration with unacceptable errors. Acceptable calibrations could be obtained by changing the geometric orientation of the camera or projector system or the lenses. The problem occurred more frequently when short focal length lenses were used, but the exact cause of the errors was not known. Recent applications of the DSL 3-D system required the use of shorter focal length lenses, and it was difficult to obtain satisfactory results even after several calibration attempts. 

Approach - This project was directed toward analysis of the problem and development of an improved calibration technique. Data from many previous calibration attempts were analyzed to identify patterns or conditions causing unacceptable results. It was determined that one type of error was caused by inaccurate spatial calibration of the camera-lens combination. Several different camera models were investigated to find a more accurate method of spatial calibration. A technique based on Zhang (Zhang, Z., "A flexible new technique for camera calibration," Technical Report MSR-TR-98-71, Microsoft Corporation, December 1998) was selected and implemented. This method uses many views of a reference target to reduce the effects of error in a single measurement. Compared with prior techniques, which rely on precise positioning and orientation, this method is easy to use, flexible, and accurate. It also provides optical distortion coefficients of the lens so that geometric corrections can be made. 

A second type of error identified is caused by noisy data used to calculate the calibration coefficients. Analysis of results from a series of configurations showed that the effect of noisy data is much greater when the radius of curvature of projected grid lines is large. Based on this condition, criteria were established for acceptable grid line curvature.

Accomplishments - An improved calibration procedure was developed and implemented. This procedure uses the Zhang technique for spatial calibration and includes tests for noise and grid line curvature to detect conditions that are likely to lead to unacceptable calibrations. The improved procedure is now being used successfully for DSL 3D measurements and systems. 

2005 Program Home