Home > Research > Validation

Validation

Overview / Introduction

As a result of collaborations and funding from our partners, VSR has grown into a center of expertise for motion-capture-based validation (experimentation and statistical analysis).  VSR has developed a validation methodology to analyze statistically the predictive dynamics of a virtual humans (Rahmatalla et al, 2008a, 2008b, 2009a, 2009b).  The objective of developing a validation methodology is to efficiently and effectively compare the motion of the predicted model and normal subjects while requiring a minimal amount of information.

Methods / Current Research

The proposed validation methodology involves testing the ability of the predicted model to pass through four benchmark tests. Figure 1 shows a flow chart of the proposed validation benchmark tests which contains qualitative and quantitative comparisons.

validation-flow-chart.gif
Figure 1: Validation flow chart

The qualitative comparison stage of validation includes the first two benchmark tests.  In Benchmark #1, a group of subjects are asked to visually evaluate the normality of the predicted motion using the VAS test. If the model passes this preliminary benchmark test, then it goes to the next benchmark test where a more detailed subjective comparison process is conducted on the kinetics and kinematics of the predicted determinants and their derivatives. 

Due to the large number of degrees of freedom in the Santos™ model and the immense amount of available information regarding the time histories for each parameter, the validation protocol identifies determinants and key frames for each task.  In this context, determinants represent the minimum number of parameters that define a given motion accurately.  Key frames are selected from the time history of the task motion in order to identify critical phases and positions with clear signatures.  For example, in the forwards walking task the flexion/extension joint angles of the lower extremities and the pelvic motion of the human are used as determinants; and the heel strike, toe off, and mid-stance positions are identified as key frames.  

Benchmark #2 is a more detailed subjective comparison process conducted on the kinetics and kinematics of the predicted determinants and their derivatives. This is done by comparing the general shape of the determinants with that of normal people in the time domain.

The quantitative comparison stage of validation includes the last two benchmark tests.  Benchmark #3 checks if the predicted motion follows the trend of the mean of the normal subjects and falls within 95 percentile confidence interval.  

With regard to statistical analysis, VSR has developed new methods of aggregating and comparing large amounts of data, in order to compare real human motion with simulated motion.  Benchmark #3 uses 95% confidence intervals to check if the predicted motion follows the trend of the mean of the normal subjects and falls within 95% confidence interval.  This analysis is shown for the six determinants of the forwards walking task in Figure 2 where the black line represents the mean of the experimental subjects, the blue dashed lines are the 95% confidence intervals and the red line is the Santos simulation.  

validation-determinants.gif
Figure 2: 95% confidence intervals are show in dashed blue lines for the walking determinants.  The experimental subject’s mean is shown in black and Santos’s determinants are shown in red.

Benchmark #4 checks if the predicted determinants can successfully pass through critical distinctive signature key-frames in the motion in a manner similar to that of normal subjects.  The coefficient of determination (R2) is used to correlate the simulation and experiment at the selected key frames, as shown in Figure 3.  The Santos simulation data is plotted on the vertical axis and the mean experimental data is plotted on the horizontal axis.  A linear fit provides an equation of the form y=mx+b where y=x indicates perfect correlation between the simulation and experimental data and deviation in the y-intercept or slope of the line provides meaningful information about the differences.  The closer the points fall to a straight line, the higher the calculated R2.  

validation-key-frames.gif
Figure 3: Selected key frames of the walking cycle, pictures from left to right: (1) Left Heel Strike, (2) Right Toe Off, (3) Left Mid-Stance, (4) Right Heel Strike, (5) Left Toe Off, (6) Right Mid-Stance; the graph shows the R2 for the six walking determinants at the selected key frames in the pictures.

The complete validation framework has been successfully tested on several tasks including forwards walking, backwards walking, sideways walking, running, box lifting, pushing/pulling, throwing, and ascending and descending stairs climbing.  Validation is ongoing for walking on an incline/decline; high crawl; and rifle aiming in standing and kneeling positions.

With regards to validating complex scenarios that involve multiple tasks, VSR has worked to develop expertise in the data collection, post processing and validation of complicated scenarios such as ingress/egress.  The ingress/egress scenario is comprised of walking, ladder climbing, and sitting tasks, with transitions connecting these tasks.  The capture volume was large, and post processing required resolution of issues with marker occlusions from the cab and subjects.  We propose applying currently available methodologies and processes in order to validate newly developed predictive dynamics simulations.

Contact Info

Salam Rahmatalla, Ph.D., Assistant Professor, Civil and Environmental Engineering, Center for Computer-Aided Design (CCAD), The University of Iowa, Iowa City, IA 52242, USA. Tel: 319-335-5614, Fax: 319-384-0542,

E-Mail: salam-rahmatalla@uiowa.edu

Related Publications

  1. S. Rahmatalla, Y. Xiang, R. Smith, J. Meusch, J. Li, T. Marler, B. Smith, “Validation of Lower-Body Posture Prediction for the Virtual Human Model SantosTM,” 09DHM-0027, SAE Sweden, 2009a.
  2. S. Rahmatalla, Y. Xiang, R. Smith, J. Meusch, J. Li, R. Bhatt, K. Abdel-Malek, “Validation of Santos Biomechanics,” Proceedings of the ASME Summer Bioengineering Conference, California, 2009b.
  3. S. Rahmatalla, Y. Xiang, J. Li, R. Smith, J. Meusch, R. Bhatt, “A Validation Methodology for Predictive Dynamics Human Models,” in review, ASME Journal of Biomechanical Engineering, 2008a.
  4. S. Rahmatalla, Y. Xiang, R. Smith, J. Li, J. Meusch, R. Bhatt, C. Swan, J.S. Arora, K. Abdel-Malek, “A Validation Protocol for Predictive Human Locomotion,” 08DHM-0024, SAE International 2008b.
  5. J. Yang, T. Marler, S. Rahmatalla, “Validation Methodology Development for Predicted Posture,” paper No. 2007-01-2467, SAE's 2007 Transactions Journal of Passenger Cars - Electronic and Electrical Systems 2007.