Classification of Anthropometric Data using Neural Networks

Ricardo Ferreira Vieira de Castro

1,2

, Pedro Henrique Gouvêa Coelho

Joaquim Augusto Pinto Rodrigues

1,2

and Luiz Biondi Neto

State University of Rio de Janeiro, FEN/DETEL, R. S. Francisco Xavier,

524/Sala 5006E, Maracanã, RJ, 20550-900, Brazil

Instituto Nacional de Tecnologia - INT, Rio de Janeiro, Brazil

Keywords: Computational Intelligence, Neural Networks, Kohonen Networks, Anthropometric Data.

Abstract: This paper proposes the use of neural networks to help with the solution to a problem demanded by the

productive sector in the manufacture of work cabinets compatible with the characteristics of a group of

individuals that characterize a good sampling of the studied population. The study is intended to serve as a

basis for further work aimed for good ergonomics in meeting the basic conditions necessary for the comfort

and welfare of the operators of these jobs. In this investigation we used Kohonen neural networks to sort the

data related to the height of the seat-eye level and length of the seat forearm-hand. The results showed that

the use of this tool is effective and allows its application in studies using more anthropometric variables

making possible to explore further needs.

1 INTRODUCTION

Anthropometry is the field of anthropology that

studies the physical dimensions of the human body.

For this reason, studies are focused on the

acquisition of data related to the size, length and

movements of the limbs (Ferreira, 1988). In

ergonomics two types of anthropometric dimensions

are found: static and dynamic. The static

corresponds to physical measurements of the body at

rest, while the dynamics are related to measures of

body in movement. To apply the data correctly, it is

important to evaluate the key influencing factors

such as race, ethnicity, diet, health, physical activity,

posture, body position, clothing, time of day etc.

(Minetti, 2002).The anthropometric Measurements

of a user serve to adapt the production means, when

using any tool or instrument. Anthropometry helps

to: evaluate positions and distances to the range of

control devices and information and to define spaces

around the body, identify objects or features that

prevent or interfere with the movement. According

to (Minetti ,2002) when the machinery or equipment

fit properly to the body, from the point of view of

dimensional errors, accidents, discomfort and

fatigue, decrease significantly. Anthropometric

methods are among the basic tools to work for the

evaluation and project development in which the

variations in size, proportions, mobility, strength and

other factors that define physically human beings are

considered. This paper is organized in five sections.

The first section is the present introduction. The

second section discusses the anthropometry. Section

three describes the model used. Section four

describes the method of data acquisition and shows

results and discussions and the paper ends with

section five depicting results and future work.

2 ANTHROPOMETRIC SURVEY

The anthropometric survey data shows the

variability of the dimensions of a population,

therefore measures that refer to a population in

another region with different socio-economic levels,

age and sex can not be taken into account (Barros,

1996). The anthropometric measures of a research

are essential data bases for designing a position that

satisfies ergonomically the employees because only

from the dimensions of individuals is that one can

define, in a rational basis, the proper sizing, both for

the working machine such as the activity involved

aiming basically, safety, efficiency and worker

comfort. The first step is then to obtain the

anthropometric measures of the operator in order to

adapt the work to the operator, in order to achieve a

116

Ferreira Vieira de Castro, R., Henrique Gouvêa Coelho, P., Augusto Pinto Rodrigues, J. and Biondi Neto, L.

Classiﬁcation of Anthropometric Data using Neural Networks.

DOI: 10.5220/0004002301160119

In Proceedings of the 14th International Conference on Enterprise Information Systems (ICEIS 2012), pages 116-119

ISBN: 978-989-8565-12-9

correct posture, a more favorable body position and

higher speed and precision of muscle movements,

thereby increasing the efficiency limb movements of

the operator. The dimensions of body parts vary

from individual to individual, but also in the same

body, throughout its life. There is no individual

whose dimensions are fully harmonic, i.e. they are

all components defined on average basis. Figure 1

shows one anthropometric measure being carried

out.

Figure 1: Horizontal distance between the top surfaces of

the lateral deltoid muscles.

3 KOHONEN NEURAL NET

As for all the neural networks, the Kohonen neural

networks are formed by a number of simple

elements, called neurons arranged in more complex

structures that work together as shown in Figure 2.

Each neuron is a processing unit that receives inputs

from outside world or from other neurons, and

produces a response to other neurons or to the

outside world. As the neurons of the brain, neural

networks are interconnected by branches through

which the stimuli are propagated. The learning

process strengthens the links that lead the system to

produce more efficient responses.

What distinguishes a Kohonen network from the

others is a two layer structure in which one is its

input and the other is for processing where a map is

formed.

Figure 2: Kohonen Neural Network Structure.

The Kohonen feature map searches the

organization of relationships between patterns. The

arriving patterns are classified by the units they

activate in the so called competitive layer.

Similarities between patterns are mapped in

proximity relationships on the grid near the

competitive layer. Once training is complete, the

relationships between patterns and clusters are

observed in the competitive layer.

The Kohonen network provides advantages over

the classical techniques of pattern recognition

because it uses the parallel architecture of a neural

network and provides a graphical organization of the

relationships.

4 RESULTS

The universe of the original study consisted of a

population comprised of 338 individuals 231of

which was men and 107 was women where 51

anthropometric variables were taken into account for

the work project and tools. All data were stored in an

Excel spreadsheet that contained information for

each operator. For each anthropometric measure,

statistic data were extracted. Figure 3 shows an

example of an anthropometric report concerning the

measurement HEIGHT OF EYE LEVEL -

SITTING.

The data statistical analysis was performed by

using percentiles which is a separator which divides

the distribution in 100 equal parts, from smallest to

largest, for any specific type of body size. The

percentiles used in the analysis yield the values for

the studied variables of 1, 2.5, 5, 25, 50, 75, 95, 97.5

and 99.

In this research work, the MATLAB software

was used including the set of routines Somtoolbox.

Practical tests were performed in order to find a SOM

(Self Organized Map) model to analyze the data

repository. Twenty five neurons was used comprising

a grid of 5x5 neurons. The input variables used were:

Height of Eye Level – Sitting, and Seated Leg-

Length Forearm as shown in Figure 4.

Through the application of a system based on a

Kohonen Neural Network which uses a

unsupervised method it was possible to make a

correlation between two measurements and a

representative of degree of relevance to the

operation and control for work cabinet development.

Figure 5 shows an example of Ergonomic Design of

of Control Room where the two anthropometric

measurements, object of this study, were taken into

account.

Classiﬁcation of Anthropometric Data using Neural Networks

117

In a previous study from the authors, statistical

results were obtained only for each anthropometric

measurement, without further identification of any

Figure 3: Sample of a Report Concerning an

Anthropometric Measurement.

Figure 4: Height of Eye Level - Sitting / Seated Leg-

Length Forearm.

Figure 5: Ergonomic Design of a Control Room.

correlation between Height of Eye Level - Sitting

and Seated Leg-Length Forearm. Using a 3 x 3

matrix Kohonen it is possible to identify groups that

have similarities in the relationship between those

two anthropometric measurements. Figure 8 shows

the results of the SAMPLE SOM HITS algorithm

applied (Kazapi, 2004).

Figure 6 shows the distribution of points in a two

dimensional plane and their respective centroids in

blue dots (Sewo, 2003). The plot depicts the

distribution regarding the two weights along the

plane after the kohonen neural network training

process.

Figure 6: Results of the Sample Som Hits Algorithm.

Figure 7: Som Weight Positions.

5 CONCLUSIONS

Nine groups of percentiles were used in the

statistical model. To develop the mathematical

model for the correlation of two anthropometric

measures under study would require a certain

expenditure of labor devoted to this task. The work

presented here explores the use of categories for

classifying patterns that in our study is very

important to identify the groups that have

similarities in content between anthropometric

measures dedicated to this study. The results here

shown are preliminary and for the completion of the

research a large data base would be needed. The

current results indicate the research seems to be in

the right direction.

REFERENCES

Minetti, L. J., Souza, A. P., Alves, J. U., Fiedler, N. C.,

2002. Anthropometric Study for Chainsaw Operators.

In Brazilian Agricultural and Environment

Engineering Magazine, Vol. 6, no. 1, pp. 166-170.

ICEIS 2012 - 14th International Conference on Enterprise Information Systems

118

Campina Grande, Pb. In Portuguese.

Sewo, J., Silva, P. R. R., 2003. Neural Network Trained in

Non Supervised Algorithm for Pattern Recognition

Purposes, Federal University of Mato Grosso do Sul,

Technical Report. In Portuguese.

Kazapi, R. G., 2004. Analysis of Using Techniques in

Pattern Recognition for Anthropometrics, Federal

University of Santa Catarina, Florianópolis, Technical

Report. In Portuguese.

Ferreira, D. M. P., Guimarães, L. B. M., and Cuiabano, A.

M. S. C., 1988. Anthropometric and Biomechanical

Research for Processing Industry workers, National

Institute of Technology, Rio de Janeiro. In Portuguese.

Classiﬁcation of Anthropometric Data using Neural Networks

119