Performance Analysis

To evaluate the performance of our registration and tracking subsystems, we have performed an extensive set of controlled perturbation studies [6]. In these studies, we have taken existing data sets, simulated data acquisition from the surface of the data, added noise to the simulated surface data, then perturbed the position of data and solved for the optimal registration. Since we know the starting point of the data, we can measure the accuracy with which the two data sets are re-registered.

While extensive details of the testing are reported in [6], the main conclusions of the analysis are:

• accurate and stable registration is achieved for up to 45 degree rotational offsets of the data sets, with other perturbations;

• accurate and stable registration is achieved for up to 75 degree rotational offsets of the data sets, with no other perturbations;

• robust registration is obtained when the surface data spans at least 40% of the full range of the surface, and is generally obtained with as little as 25% coverage;

• small numbers of outliers do not affect the registration process.