Content area
Abstract
This research focused on error reduction in RSA through the introduction of a non-linear reconstruction algorithm and distortion correction polynomials, along with a novel calibration frame design. Repeated measure tests were conducted with a custom gauge object. Evaluation criteria included measurement of accuracy, precision and robustness.
An improved level of accuracy, precision and sensitivity to noise were provided with a non-linear bundle adjustment (BAM) solution to the collinearity equations versus the standard direct linear transformation (DLT) reconstruction algorithm. The addition of affine distortion corrections significantly improved accuracy. The addition of radial distortion correction was not supported for either analog or digital films. The addition of decentric distortion correction provided statistically significant improvements with digital films, but not analog films. The novel multiplanar calibration frame provided statistically significant improvements to reconstruction error relative to the traditional uniplanar type calibration frame. RSA levels of accuracy and precision were superior when data was acquired with digital rather than traditional anolog film formats.
The ability of several optimization schemes to provide accurate and precise 3D reconstruction of a poorly calibrated calibration frame was quantified. The RSA precision provided with reconstruction algorithms was assessed clinically on double examinations of five patients. DLT based reconstruction algorithms provided fewer mathematical convergences and appeared less reliable than their BAM counterparts. No statistically significant differences were observed between the RSA precisions provided with the BAM and DLT reconstruction algorithms in the clinical data. Future work should focus on extending this analysis with a larger clinical sample size, incorporating the multiplanar calibration frame.
