100 Mbps switched Fast Ethernet network. Each machine is separately seeded with a

random set of initial patient-specific model parameter values. The outer-level

optimization program was implemented in C on the Linux operating system with the

Message Passing Interface (MPI) parallel computation libraries.

 Two-Level Optimization Evaluation

Synthetic Marker Data without Noise

 To evaluate the ability of the two-level optimization approach (Figure 3-10) to

calibrate the generic, parametric kinematic model, synthetic movement data was

generated for the ankle, knee, and hip joints based on estimated in vivo model parameters

and experimental movement data. For each generated motion, the distal segment moved

within the physiological range of motion and exercised each DOF for the joint. There

were 50 time frames and approximately 3.5 cycles of a circumductive hip motion

consisting of concurrent flexion-extension and abduction-adduction. Flexion-extension

comprised 50 time frames and roughly 4 cycles of knee motion. The ankle motion

involved 50 time frames and nearly 2.75 cycles of circumduction of the toe tip, where

plantarflexion-dorsiflexion and inversion-eversion occurred simultaneously. The ability

of the two-level optimization to recover the original model parameters used when

generating the synthetic motions was assessed.

Synthetic Marker Data with Noise

 To evaluate the ability of the two-level optimization method (Figure 3-10) to

calibrate the generic kinematic model to a synthetic patient, skin movement artifacts were

introduced into the synthetic movement data for the ankle, knee, and hip joints. The

relative movement between skin and underlying bone occurs in a continuous rather than a

random fashion (Cappozzo et al., 1993). Comparable to the simulated skin movement