The purpose of rigid-body registration is to find a rotation and translation which transforms one coordinate frame to another. The procedure requires measurement of common points (fiducials) in both frames. Noise and possible bias in the acquired locations of fiducials adversely affect the quality of registration. While the influence of noise can be mitigated by taking repeated measurements, bias remains in the mean locations of fiducials and distorts the rigid-body condition. We present a numerical method which allows for the compensation of bias in the locations of fiducials and the restoration of the rigid-body condition. Once the registration transformation is calculated, it can be used to transform other points which are measured only in one coordinate frame (target points). We estimated the bias in these points by linearly interpolating the biases from neighboring fiducials. After removal of the bias from the target location, there was a decrease in the target registration error. This method was evaluated using three different datasets acquired with three different instruments.
NIST Interagency/Internal Report (NISTIR) - 8180
bias, noise, point-based rigid-body registration, target registration error