Introduction

For Human Body Model (HBM) scaling, example of scenarios include users that would like to generate the most likely subject matching a few characteristics:

a given height and weight (e.g. average female)
one of the height, seating height and shoulder width specified in the EC Regulation R129 for child restraint systems
the few measurements available on a PMHS in a published validation study
...

Alone, such characteristics may be insufficient to drive the scaling of a HBM. However, they could be completed by using statistical relationships (a priori knowledge) linking these known characteristics (called predictors) to outputs such as anthropometric dimensions or shapes. The anthropometric dimensions can then be used in other modules such as the Scaling Constraint Module to scale a model. The current module focuses on the prediction of anthropometric dimensions based on a limited but arbitrary set of predictors.

Description / methodology

The Anthropometric Prediction module can be used to generate anthropometric targets (in the sense of the PIPER Framework) based on a set of predictors and anthropometric datasets. The underlying statistical process is based on Parkinson and Reed 2010 (see References). It is fully automated.

A regression between selected predictors, such as weight, is first evaluated for a chosen population. A set of targets is then sampled from this regression measurements that can be used in other modules to personalize a model.

Targets can be sampled from the regression by including the regression error or not (sampledOutput or meanOnly). Similarly, the values of the predictors can either be forced to a given value or sampled from a statistical distribution (Fixed Predictor/Sampled Predictor). That statistical distribution is estimated from a population that can be the same as the one used to generate the regression or from a different one.

Populations are constructed by upsampling through the definition of some bins or subsets of population they consist of. Each bin is defined by a set of criteria. For example, a population could contain a single bin consisting of female subjects between the ages of 20 and 35, or it could consist of two bins of female subjects respectively between the ages of 18 and 35, and the ages of 36 and 65. Please note that you have to define the same amount of criterias in each bins.

The module can deal with multiple datasets. Three publicly released datasets are provided with the release and named as follows:

ANSUR: public dataset with anthropometric measurements on military personnel (almost 4000 subjects, ages: 17-51. See References)
CCTANTHRO: dataset released under a CC-BY 4.0 Licence with anthropometric measurements on Post Mortem Human Subjects (105 subjects, ages: 55-94. See References)
SNYDER: public data corresponds to anthropometric measurements performed on children (about 3900 subjects, ages: 1-19, not all measurements available for all subjects. See References)

Remarks: Warning: while the process is automated, it is critical to remember the principles of the underlying process. The plausibility of the predicted target is widely dependent on the relationship between the set of predictors and the dataset (how far is the set of predictor from actual data present in the dataset). Depending on the selection of predictors and their value, the predicted target may be based on unrealistic extrapolations. More feedback about this will be added in the future.

Parameters and Options

Parameters :
Optional: none
Inputs :
Interactive menus (opening on the right of the window) help the user selecting the inputs. The set of anthropometric targets are then generated and saved in the results directory by clicking on the "Generate Targets" button. The target can be generated when all indicators are in green (all inputs ok)
Parameters :
Select if you want to generate a New Regression of reuse one that was previously generated and saved.
Dataset options:
- Select the dataset (see Description / methodology)
- Select the variables that will be generate (=target or measure of interest) and predictors. For SNYDER: not all measurements are available for all subjects. For ANSUR: all measurements are generated.
- Select the population characteristics: this is the population you want to base the target generation on. First we need to determine how many bins/subsets of population we want to work with. Once you have defined the number of bins, each bin can be defined with this optional set of criteria: % of population, gender, ethnicity, age
  - % of population : Set the percentage of the population described by this bin.
  - Gender : Set the gender of this bin of population.
  - Ethnicity : Set the ethnicity of this bin of population.
  - Age : Set the age range of this bin of population (pre-filled with the maximum range)
Target information
- Sample Type:
  Here you can select if you want the sample type to be meanOnly type, sampledOutput type or both. If you desire mean values of the anthropometric measurements for specific values of the predictors, choose 'meanOnly'. Otherwise, samples will be drawn among the subjects that share the same predictor values and you will select "Normal" for "sample output type", to indicate that the samples use normal distribution.
- Sample input type:
  Here you decide if the values of the predictors you previsously selected in the regression definition, will be sampled or fixed predictor.
  - Fixed Predictor
    The choice of Fixed Predicators allow you to determine an exact value for each one of the predictors you selected from the predictor list. If you chose "sampledOutput" as sample type, you will have to determine the number of samples of sets of targets you desire.
  - Sampled Predictor
    - Sample input type Select the type of sample input, only "normal" available
    - Number of samples Define the number of sampled predictors that will be used for the target generation
    - sample Input reference population Mat file Define the input reference population mat file that may be used when the statistical distribution is estimated from a population different from the one used in the current regression (upsampled populations are saved in the regression *.mat files together with the regression information).
- Sample Output type Select the type of sample output, only "normal" available
Output options
- Select where and under which name the target files will be saved. The files will be numbered automatically when there are several
- (optional) Select where the regression will be saved if you would like to re-use it later.

Remarks: The functionalities are implemented into a set of Octave scripts called by a GUI in the PIPER application. Some parameters are not implemented in the GUI and are available in the code.

References

ANSUR: Gordon CC, Bradtmiller B, Churchill T, Clauser CE, McConille JT, Tebbetts I, Walker RA. 1988 anthropometric survey of U.S. army personnel: methods and summary statistics. Technical Report NATICK TR-89 044. Natick, MA: U.S. Army Natick Research, Development, and Engineering Center. 1989. Data downloaded from http://mreed.umtri.umich.edu/mreed/
CCTantrho: database released under CC-BY-4.0 by CEESAR in 2017 and distributed by PIPER. Publication pending.
Parkinson, M. and Reed, M.P. Creating virtual user populations by analysis of anthropometric data. International Journal of Industrial Ergonomics, 2010, 40, 106–111
SNYDER: Snyder, R, L Schneider, C Owings, H Reynolds, D Golomb, M Sckork. (1977) Anthropometry of infants, children, and youths to age 18 for product safety design. UM-HSRI-77-17, CPSC, Bethesda, MD. Data downloaded from http://mreed.umtri.umich.edu/mreed/