A method for body posture classification of three-dimensional body models in the sagittal plane

Abstract

The main purpose of this research was to develop a method for the classification of body posture types, using a three-dimensional body scanner and data on anthropometric measurement. A sample of 102 male test subjects, aged from 20 to 30 years, and without structural deformities of the locomotor system, were scanned. Anthropometric body measurement was performed and 16 measurements were selected to calculate upper and lower body curve angles as a set of posture indicators. The number of maximally different groups of test subjects was determined using k-means cluster analysis and factor analysis of established posture indicators. The sampling into three upper and three lower posture types showed the largest statistically significant differentiation, comparing the results of variance analysis between and within the obtained groups, which confirmed three main components extracted by factor analysis. Discriminant analysis used in order to determine differences between posture types showed which indicators are the most important when classifying test subjects belonging to a particular upper or lower posture type. Classification functions were defined based on discrimination functions and their factor loadings, and used to calculate a matrix of correctly classified test subjects. The classification matrix showed very high prediction possibility of established posture indicators and the proposed method for body posture classification.

Keywords

body posture classification method three-dimensional body scanner sagittal plane garment construction

The problems of body posture are in the focus of researchers and experts from various fields. Given the importance of body posture for every individual, but also the consequences resulting from improper body posture, the subject is primarily of concern to researchers from the fields of medical sciences, kinesiology, and anthropology.^1–5

Under the concept of good or correct body posture, different definitions exist in the literature. Garrison and Read⁶ reported that good posture involves proper alignment of body segments and their balance, achieved with minimal power investment and maximal mechanical efficiency. According to Welket al.,⁷ the correct posture of the human body is an upright position with relaxed arms at the side and palms pointed forward. Characteristics of correct body posture can be seen from the anteroposterior and sagittal views. Viewed from the front side, all individual body parts should be in perfect symmetry, and in the sagittal view all segments should be aligned with the gravity line positioned through the center of the body. Most of the authors describe four body posture types: excellent, good, poor, and irregular posture.^8,9

Diagnostic methods used to analyze body posture and the presence of functional disorders or deviations of particular body segments can be performed in several ways and are characteristic for specific fields of application.^10–12 A large number of methods, in the medical field, involve applying sophisticated equipment that can diagnose changes in the locomotion system at the very early stage of formation. However, these methods are not applicable for a wide range of uses and some of them are invasive, such as x-ray radiation.¹² Methods applied in the field of kinesiology are more acceptable for serial measurements of a larger set of test subject samples, especially children and youth where the preventive control of posture is performed at a different stage of growth.^13,14

The method developed by Barrow and Mec Gi¹⁵ for the examination of school-age children, can be pointed out among the major methods for body posture assessment in the sagittal plane. The method is based on a comparison of body parts in the lateral and in the anteroposterior position with a particular card representing each of 13 body parts. The measurement instrument Scoliosometer, developed by Tribastone,¹⁶ enables the determination of various posture indicators by placing a test subject in front of a plexiglass board with a drawn grid mesh used for analysis of distances and deviations of indicators in the frontal and sagittal views. This measurement instrument can identify any type of asymmetry or deviations of orthostatic posture.

A number of methods are based on photographing test subjects in a defined position and computer assessment of posture based on the two-dimensional (2D) body image.¹⁷ The method developed by McEvoy and Grimmer¹⁸ uses such computer-based image processing for determination of posture indicators. A 2D body image of a test subject is captured in the frontal and in sagittal positions with markers placed on the selected referent body points. The image is further processed to determine the deviation angles of the left and right posture indicators in the frontal plane and deviations from the gravity line in the sagittal plane. The method developed by Pausic¹⁹ involves a new measurement instrument used for the assessment of body posture on school-age boys. The new measurement instrument differs in three body posture types in the sagittal view, on boys aged 10–13. All the presented methods differ in the method of posture assessment and the selection of posture indicators, based on the application for which they are developed, and there is no standardized procedure for posture analysis.

The issue of body posture is also one of the important topics in clothing design and construction, given that it greatly affects garment drape on the human body, especially in the sagittal view. Analysis of different body shapes, along with identification of different morphologies, helps us to recognize the importance of adapting new measurement charts for different fit specifications.²⁰ Most such applications mainly classify body figure types. The morphotype analyzer developed by Human Solutions enables the classification of body figure types and their newest methodology includes dividing analysis into the upper and lower body. In this way they covered greater variability of human bodies. Also, they are very precise in figure type analysis, but the posture classification includes only three possible types, Z, S, and I, which is very limited for made-to-measure clothing development.²¹ Body posture is specified with bone size and structure and represented with the shape and place of the front and back body curve in the sagittal view.²⁰ While many existing algorithms have been focused only on detecting body posture irregularities, some algorithms are able to classify a given body posture into the one of several predefined categories.¹⁹ These posture assessment algorithms are quite valuable to the fashion and garment industry because the presence of body trunk functional changes greatly affects the garment construction method and requires certain modifications and pattern adjustment.^22–24 In that sense, body posture assessment can be related to clothing construction by using the determined values of deviations from normal posture as construction modification rules or by direct mapping of the back middle line as a segment of the cutting pattern (Figure 1²⁴).^23–28

Fashion designers and manufacturers are increasingly relying on computer-based simulations especially when designing functional clothing with added value, such as military and working uniforms, high-quality business clothes, sports clothes, etc.^29–31 Body posture is critical for accurate computer-based simulation/prediction of garment drape on the human body.²² Computer-aided design (CAD) systems for clothing simulations usually use parametric body models that can be customized according to individual measurements, which allows designers to visualize how the clothing is likely to drape on different types of bodies. One of the main insufficiencies of parametric body models is a lack of posture adjustment. This is why scanned body models of targeted individuals are used when designing made-to-measure clothes of targeted fit.³²

The application of modern technologies, such as the three-dimensional (3D) body scanner, allows a detailed analysis of anthropometric characteristics, relationships of body proportions and analysis of body posture and symmetry on individual test subjects.^31–33 The standardized procedure of the automatic computer-based body measurement is used to take 150 body measurements that are primarily used for garment construction and pattern modification according to individual measurements.³⁴ However, there is no standardized method for body posture assessment and posture analysis is performed by visualizing a computer-based 3D body model in basic anatomical planes or the plane defined by the examiner, while specific body posture indicators are not defined.

In that sense, the presented research was performed with the primary objective to define a method for body posture classification that will be valid, reliable and objective. According to this purpose, a research hypothesis has been set up: the application of a 3D body scanner and systematic selection of anthropometric measurements for calculation of posture indicators in the sagittal plane enables one to define a method for classification of various body posture types, which significantly differ in indicator space. The second objective of the research was to determine and define the exact description of a particular posture type based on the determined ranges of values of body posture indicators. The third research objective was to define a sample structure in terms of the test subject presence in the particular body posture type.

Test subjects and methods

Sample of test subjects and measurement equipment

A sample of 102 male test subjects, aged 20–30 years, was processed in the study. The test subjects are classified as having a slim figure in size range from 44 to 60 according to ISO 3636 (Size designation of men’s clothing) included in ISO 8559-2:2017. A standard defines the male slim figure in terms of a difference of at least 14 cm between chest and waist circumference and 16 cm difference between waist and hip circumference. Scanning was performed by using a laser 3D body scanner, according to standard ISO 20685:2010, which has the purpose to ensure comparability of body measurements currently determined by standards ISO 7250:1996 (Basic body measurements for technological design) and ISO 8559:1989 (Clothing construction and anthropometric research—body dimensions), measured using a 3D body scanner and not conventional anthropometric instruments.^34–36 The standard implies scanning of subjects in upper standing A position with chin parallel to the floor and arms slightly bend at the elbow, wearing light colored, tight underwear, and a swimming cap, without any accessories. In addition, test subjects were standing on precisely positioned floor marks to ensure that every test subject was standing in the same position. The automatic measurement procedure, according to the standard, includes set of 150 body measurements. For our study, we created a customized measurement menu with the selection of measurements for posture analysis. The customized measurement menu includes basic body circumference measurements of the bust, waist, and hips, and 14 measurements for calculation of posture variables. Descriptive statistics and basic distribution parameters of the main body measurements by garment sizes are presented in Table 1. Garment sizes are defined by main body circumferences, including bust girth (BG), waist girth (WG), and hip girth (HG).

Table 1.

Descriptive statistics and distribution parameters of main body measurements by size

Body measure	Garment size	Percentiles
Body measure	Garment size	N	Mean [cm]	SD	CV%	Min [cm]	10th [cm]	90th [cm]	Max [cm]
BG	44	9	88.8	0.6	0.7	87.8	87.8	89.6	89.6
	46	22	92.3	1.0	1.1	90.2	90.7	93.5	94.0
	48	23	96.2	1.2	1.2	94.2	94.5	97.5	97.8
	50	19	99.8	1.1	1.1	98.2	98.4	101.5	102.0
	52	10	103.7	1.1	1.1	102.1	102.1	105.1	105.3
	54	10	107.9	1.2	1.1	106.2	106.2	109.4	109.7
	56	3	111.8	1.1	1.0	110.4	110.4	113.1	113.1
	58	4	115.2	0.4	0.3	114.7	114.7	115.7	115.7
	60	2	123.7	3.4	2.7	120.3	120.3	127.0	127.0
WG	44	9	73.8	1.9	2.6	70.1	70.1	75.8	77.1
	46	22	73.9	2.8	3.8	68.6	69.5	76.5	79.6
	48	23	77.5	3.1	4.0	73,.7	73.8	82.8	83.6
	50	19	81.1	2.9	3.6	76.4	76.5	84.6	87.3
	52	10	82.9	4.4	5.3	76.6	76.6	86.4	92.0
	54	10	87.7	5.0	5.7	80.9	80.9	94.0	94.0
	56	3	92.3	3.2	3.4	89.5	89.5	96.7	96.7
	58	4	96.5	5.7	5.9	88.3	88.3	102.1	102.1
	60	2	106.9	2.2	2.0	104.7	104.7	109.0	109.0
HG	44	9	94.9	3.2	3.4	92.2	92.2	96.9	102.8
	46	22	93.2	3.4	3.6	87.4	88.0	96.8	99.6
	48	23	97.1	2.1	2.1	93.1	94.5	99.2	101.6
	50	19	100.4	4.5	4.5	92.6	93.4	105.7	107.9
	52	10	100.8	4.0	3.9	94.6	94.6	106.0	107.6
	54	10	102.9	4.0	3.9	94.7	94.7	108.5	108.5
	56	3	104.3	4.0	3.8	99.3	99.3	109.1	109.1
	58	4	109.6	3.7	3.4	105.6	105.6	115.3	115.3
	60	2	110.9	2.4	2.1	108.5	108.5	113.2	113.2

BG: bust girth; WG: waist girth; HG: hip girth.

Variables

The size, shape, and position of bone segments presented as a skeleton structure have the most impact on body posture, forming the shape of the front and back body curve in the sagittal view. Longitudinal and transversal body cross-sections were analyzed on targeted anthropometric points (Figure 2). Based on cross-sections analysis, 14 measurements on a 3D body model were selected from the set of measurements, in order to describe body curves and define posture indicators for further analysis (Figure 2). We selected eight length measurements defining the distance of anthropometric points on the back middle line from the posterior vertical plane in the sagittal view: the distance of the seventh cervical vertebra from the vertical posterior plane (c7); distance of the most prominent point on the scapula from the vertical posterior plane (Scp); distance of the back middle point on the chest line from the vertical posterior plane (CB); distance of the back middle point on the waist line from the vertical posterior plane (WB); distance of the back middle point on the hip line from the vertical posterior plane (HB); distance of the front middle point on the chest line from the vertical posterior plane (CF); distance of the front most prominent point on the belly line from the vertical posterior plane (ByF); and distance of the front most prominent point on the abdomen line from the vertical posterior plane (AbF). Six selected height measurements are define height distances between selected points: height of the seventh cervical vertebra (c7H): height of the most prominent scapula point (ScpH); chest height (CH); waist height (WH); hip height (HH); and belly height (ByH) (Figure 2).

Using the determined measurements, the angles between planes positioned through characteristic anthropometric points were calculated and defined as posture indicators, given that the obtained angles approximate for the front and back body curve in 3D space (Figure 3). In this way, the number of necessary variables for posture analysis was reduced and the specific angle values can be further used in clothing construction for pattern customization. Targeted curve angles are defined as the tangent of the ratio between the difference in distance from the vertical posterior plane and the height of the correspondent anthropometric points (Figure 3). In this way, six variables were defined as posture indicators for further analysis and classification.

Posture indicator Q1 represents the angle of the back middle curve between the seventh cervical vertebra and chest line, Q2 represents the angle of the back middle curve between the chest and waist line, Q3 represents the angle of the back middle curve between the waist and hip line, Q4 represents the front curve angle between the chest and belly line, Q5 represents the front curve angle between the belly and hip line, and Q6 represents the angle between the seventh cervical vertebra and the most prominent scapula point (1–6)

Q 1 = {tan}^{- 1} \frac{Δ c 7 - CB}{Δ c 7 H - CH} [^{\circ}]

(1)

Q 2 = {tan}^{- 1} \frac{Δ CB - WB}{Δ CH - WH} [^{\circ}]

(2)

Q 3 = {tan}^{- 1} \frac{Δ WB - HB}{Δ WH - HH} [^{\circ}]

(3)

Q 4 = {tan}^{- 1} \frac{Δ CF - ByF}{Δ CH - ByH} [^{\circ}]

(4)

Q 5 = {tan}^{- 1} \frac{Δ ByF - AF}{Δ ByH - HH} [^{\circ}]

(5)

Q 6 = {tan}^{- 1} \frac{Δ c 7 - Scp}{Δ c 7 H - ScpH} [^{\circ}]

(6)

Data processing methods

The structure of the variables was analyzed using the Factor and Principal Component Analysis (PCA) method, and varimax rotation was used in order to segment the variables into significant components. The main component of this research was to develop the method for posture type classification using selected 3D scanning data as indicators. The structure of the subjects was examined using the k-means cluster analysis method in order to determine posture types, that is, define the number of most various groups of test subjects. Since the method enables self-defining the desired number of groups, we performed analysis with two, three, and four possible groups, considering the previous extraction of three main components by factor analysis. Sampling with the greatest statistically significant differentiation between groups was confirmed by comparing the results of variance analysis between groups in every posture indicator for different numbers of clusters. Posture indicators that have the most significant impact on the sampling into clusters are defined based on determined F-values and significance level p. Every cluster is presented by statistic descriptive parameters and basic distribution parameters of posture indicators. Euclidian distances of subjects in multidimensional space were determined as measures of precision, according to expression (7)

d (x, y) = {\sum_{i} (x_{i} - y_{i})^{2}}^{\frac{1}{2}}

(7)

Discriminant analysis was used to determine which variables discriminate the best between obtained clusters. Discriminant function analysis is basically multivariate analysis of variance (MANOVA) in reverse; it determines which variable differs the clusters based on the mean value of the variable in the cluster and uses that variable for prediction of cluster belonging. Eigenvalues of discriminant functions (λ), canonical correlation coefficients (R_C) and Wilks’ lambda of discriminant functions ( $W λ$ ) (8) were calculated and used to interpret the influence of a particular discriminant function on belonging to a particular cluster

W λ = | \frac{S_{wg}}{S_{bg} + S_{wg}} |

(8)

A functions structure matrix was calculated in order to determine which particular variable correlates the best with each discriminant function. The number and percent of test subjects correctly classified to the particular cluster was determined using a classification matrix calculated based on the classification function, according to expression (9)

Si = Ci + {Wi}_{1} \cdot x_{1} + {Wi}_{2} \cdot x_{2} + \dots + {Wi}_{n} \cdot x_{n}

(9)

where i is the observed cluster, 1–n are variables, Ci is the cluster constant, W is the weight of the variable, x is the observed value of the random variable, and Si is the classification score.^37–39 All results were obtained by processing experimental data in software package Statistica 12.0.

Results and discussion

The analysis was performed using six angles as indicators to describe front and back body curves and determine posture types. Analysis was divided into two directions, classification of upper body posture types represented by indicators Q1, Q6, Q2, and Q4 and classification of lower body posture types represented by indicators Q2, Q3, and Q5. Variable Q2 describes waist zone curvature. Since the waist zone is in the central spinal area, which connects the upper and lower torso, we used the Q2 variable in both upper and lower posture studies. Also, both upper and lower spinal and posture deformities are medically diagnosed in accordance to waist curvature. Thus, it was very important to include the Q2 variable in both classifications. Since it complicates the mathematical approach, we divided our posture classification method into two steps, where the first step is classification of upper posture and the second is the lower posture type, so the Q2 variable is entering analysis always within a particular targeted set of posture variables and there is no overlapping of variables. Frequency distributions of posture indicators are presented in Figure 4.

PCA and Factor analysis were performed to determine the number of components for further classification. Based on Kaiser’s eigenvalues criterion, the percentage of variance criterion and the Scree test criterion, three factor solutions were retained for further analysis of both upper and lower body posture types.

Analysis of upper body posture types

Given by the results of the upper body posture analysis, the eigenvalue for the first factor is equal to 2.011 and the proportion of variance accounted for by the first factor is approximately 50.28%, the second factor accounts for about 30.43% of the variance, and the third factor accounts for about 15.01% of the variance, giving a sum of 95.73% of total variability explained by the three extracted main components (Table 2).

Table 2.

The rotated matrix of factor loadings of upper posture type indicators

Indicator	Factor 1	Factor 2	Factor 3	Communalities
Q1	0.949	0.015	0.118	0.916
Q6	0.941	0.017	0.167	0.913
Q2	0.228	0.987	−0.158	0.999
Q4	0.167	−0.167	0.972	0.999
Eigenvalue	2.011	1.217	0.601
Variance %	50.28	30.43	15.01
The cum. variance %	50.28	80.71	95.73

Note: bold text represent statistical significance i.e. p value <0.05

Factor loadings were calculated to interpret correlations between the factors and the variables. According to communality values all variables are well represented. Substantial loadings on the first factor appear for indicators related to the back body curvature from the seventh cervical vertebra to the chest line, Q1 and Q6. Factor 2 shows substantial loading for indicator Q2 related to the back body curvature from the chest to waist line. Factor 3 shows substantial loading for indicator Q4 related to the front upper body curve. Highlighted are variables with factor loadings greater than 0.7 (Table 2).

Within cluster analysis we observed categorization in two, three and four clusters in order to find the solution with the most homogenous groups. Analysis of variance between and within obtained clusters is shown in Table 3, as well as the significance (p) of indicator mean values, showed three cluster solution as best for differentiation between subject groups, confirming the three main components exhausted by PCA.

Table 3.

Variance analysis of sample categorization in two and three clusters

No. of clusters	Sum of squares	Upper posture type indicators
No. of clusters	Sum of squares	Q1	Q6	Q2	Q4
2	Between SS	587.170	616.873	13.346	418.798
	Within SS	594.158	616.042	1302.23	926.207
	F	98.824	100.135	1.025	45.217
	p	0.000	0.000	0.314	0.000
3	Between SS	697.487	631.863	319.942	689.179
	Within SS	483.841	601.052	995.638	655.826
	F	71.357	52.037	15.907	52.017
	p	0.000	0.000	0.000	0.000

Note: bold text represent statistical significance i.e. p value <0.05

Figure 5 shows the results of sampling into three clusters with performed descriptive statistics including the number of subjects (N), arithmetical mean, minimum (min), maximum (max), and multiple percentile values adopted to quantify the sizing classification of posture types.

The three determined upper body posture types are as follows: upper posture type 1 (UP1), which describes 31 subjects representing 30.39% of total sample; upper posture type 2 (UP2), which describes 42 subjects representing 41.18% of total sample; and upper posture type 3 (UP3), which describes 29 subjects representing 28.43% of total sample.

The calculation of Euclidean distances between three upper posture types, presented in Table 4, showed significant differences between all cluster combinations.

Table 4.

Euclidean distances between cluster centroids of upper body posture types

	UP1	UP2	UP3
UP1	–	–	–
UP2	3.370	–	–
UP3	4.056	4.967	–

Discriminant analysis was used to determine differences between the obtained clusters and indicators that are most significant for belonging to a particular group. Results showed that two discriminant functions are differentiating three upper body posture types. According to values of canonical coefficients of discrimination (R_C), the first discriminant function (DF1) is most important for the differentiation between body posture types (0.835), followed by the second discriminant function (DF2) with a lower R_C value (0.651) (Table 5).

Table 5.

Discriminant analysis of upper body posture types

	Eigenvalue	R_C	Wilks’ Lambda	$χ^{2}$	df	p
DF1	2.297	0.835	0.175	170.056	8	0.000
DF2	0.735	0.651	0.576	53.733	3	0.000

According to the plot of centroids of upper body posture types for test subjects in the coordinate system of the first and second discriminant function, the first discriminant function significantly differentiates upper body curvature types UP2 and UP3, while the second discriminant function mostly differentiates type UP1 (Figure 6).

The structure of discriminant functions and centroids of the determined upper body posture types are presented in Table 6.

Table 6.

The structure of discriminant functions – correlation of indicators with discriminant functions and centroids of upper body posture types

	DF1	DF2
Q1	0.536	0.494
Q6	0.374	0.014
Q2	0.006	0.615
Q4	0.634	−0.560
Eigenvalue	2.297	0.735
Cum. prop	0.758	1.000
UP1	−0.012	1.278
UP2	1.492	−0.554
UP3	−2.149	−0.564

Note: bold text represent statistical significance i.e. p value <0.05

The correlation of body posture indicators with the first discriminant function showed that indicator Q4 (0.634) contributes mostly to the difference between the determined clusters, and the second most important indicator is Q1 (0.536). The value of the cumulative proportion shows that the first discriminant function explains 75.8% of differences between upper posture types. Given that the first discriminant function differs the best with types UP2 and UP3, it can be concluded that indicators Q4 and Q1 are responsible for belonging to those particular types. The structure of the second discriminant function showed that indicator Q2 (0.615) contributes mostly to the difference between the determined body posture types, which according to the centroids plot indicates that indicator Q2 is responsible for belonging to type UP1. Based on the cumulative proportion of explained variance it is apparent that 75.8% of all discriminatory power is explained by the first discriminant function.

Considering the obtained results, it can be established that the proposed method significantly differs for the three upper body posture types. Figure 7 shows scanned computer-based body models closest to the centroid of a particular upper body posture type according to posture indicator values. Upper posture type UP1 is best classified by the second discriminant function represented mostly with indicators Q2 and Q4 (Table 6). As shown in Figure 7, UP1 has a significantly larger body curve angle from the chest to waist line (pronounced Q2) with a noticeable angle from the seventh cervical vertebra to the chest line (Q1). Types UP2 and UP3 are best classified by the first discriminant function represented with two indicators Q1 and Q4. UP2 has a specific straight front body line (Q4 around zero) and noticeable curve from the seventh cervical vertebra to the chest line (Q1) as type UP1. Type UP3 has a significantly less severe back body curve (smaller values of Q1 and Q2) with an angled front body line specific also to type UP1 (prominent Q4).

The number and percentage of correctly classified test subjects in a particular body posture type is determined using the classification matrix (Table 7). The highest percentage of correctly classified test subjects, that is, complete prediction of 100%, was determined for upper body posture types UP2 and UP3, followed by UP1 with 96.77% correctly classified test subjects. Although the classification matrix showed a different accuracy of predicting upper posture type UP1, a total of 99.02% correctly classified test subjects indicates a very high prediction possibility of test subjects belonging to the particular upper body posture type.

Table 7.

Classification matrix for upper body posture types

Upper posture type	Correctly classified test subjects [%]	Correctly classified test subjects [N]
Upper posture type	Correctly classified test subjects [%]	UP2 p = 0.412	UP3 p = 0.284	UP1 p = 0.304
UP2	100.00	42	0	0
UP3	100.00	0	29	0
UP1	96.77	1	0	30
Total	99.02	43	29	30

Note: bold text represent statistical significance i.e. p value <0.05

Analysis of lower body posture types

Factor analysis of lower body posture type indicators showed that 100.00% of total sample variability can be explained by the three main components (Table 8). The calculated correlation between the factors and the variables showed substantial loadings on the first factor for indicator Q2 related to the back body curvature from the chest to the waist line. Factor 2 shows substantial loading for indicator Q5 related to the front lower body curvature from the waist to the hip line and Factor 3 shows substantial loading for indicator Q3 related to the lower back body curvature from the waist to the hip line (Table 8).

Table 8.

The rotated matrix of factor loadings of lower posture indicators

Indicator	Factor 1	Factor 2	Factor 3	Communalities
Q2	0.981	0.099	0.168	1.000
Q3	0.172	0.161	0.972	1.000
Q5	0.099	0.983	0.156	1.000
Eigenvalue	1.600	0.780	0.619
Variance %	53.34	26.01	20.65
The cum. variance %	53.34	79.35	10.00

Note: bold text represent statistical significance i.e. p value <0.05

The analysis of variance between and within sampling into two and three clusters showed the three cluster solution as the best for differentiation between subject groups, confirming the three main components exhausted by PCA. Figure 8 shows the results of sampling into three clusters with the performed descriptive statistics and distribution parameters. The three determined lower body posture types are as follows: lower posture type 1 (LP1), which describes 31 subjects representing 30.39% of the total sample; lower posture type 2 (LP2), which describes 39 subjects representing 38.24% of the total sample; and lower posture type 3 (LP3), which describes 32 subjects representing 31.37% of the total sample. Calculated Euclidean distances between the three determined lower posture types showed a high significant difference between all cluster combinations.

Discriminant analysis determined two discriminant functions differentiating the three lower body posture types (Table 9).

Table 9.

Discriminant analysis of lower posture types indicators

	Eigenvalue	R_C	Wilks’ Lambda	$χ^{2}$	df	p
DF1	2.537	0.847	0.159	180.018	6	0.000
DF2	0.775	0.661	0.563	56.227	2	0.000

Centroids of test subject groups of lower body posture types in the coordinate system of the first and second discriminant function show that the first discriminant function significantly differentiates lower body posture types LP2 and LP3, while the second discriminant function differentiates type LP1 (Figure 9).

The structure of discriminant functions and centroids of the determined lower body posture types are presented in Table 10. The correlation of lower body posture indicators with the first discriminant function showed that indicator Q5 (0.928) contributes mostly to the difference between the determined clusters. Given that first function differentiates types LP2 and LP3, it can be concluded that indicators Q5 is responsible for belonging to that type. The structure of the second discriminant function showed that indicators Q2 (−0.832) and Q3 (−0.464) contribute mostly to the difference between the determined body posture types, which according to the centroids plot indicates that indicators Q2 and Q3 are most responsible for belonging to type LP1. Based on the cumulative proportion of explained variance it is apparent that 76.6% of all discriminatory power is explained by the first discriminant function.

Table 10.

The structure of discriminant functions correlation of indicators with discriminant functions and centroids of lower body posture types

	DF1	DF2
Q2	0.181	−0.832
Q3	0.213	−0.464
Q5	0.928	0.365
Eigenvalue	2.537	0.775
Cum. prop	0.766	1.000
LP1	0.211	−1.307
LP2	−1.789	0.487
LP3	1.976	0.673

Note: bold text represent statistical significance i.e. p value <0.05

Considering the obtained results, it can be established that the proposed method significantly differs for the three lower body posture types. Scanned computer-based body models closest to the centroid of a particular lower body posture type according to posture indicator values are presented in Figure 10.

Lower posture type LP1 is best classified by the second discriminant function represented mostly with indicators Q2 and Q3 (Table 10). As shown in Figure 10, LP1 has significant body curve angles from the chest to the waist line (pronounced Q2) and from the waist to the hip line (pronounced Q3), with noticeable lower front body curve Q5. Types LP2 and LP3 are best classified by the first discriminant function represented with front posture indicator Q5. LP2 has a specific straight front body line (small value of Q5 angle) and less severe back body curve (small values of Q2 and Q3). Type LP3 also has a moderate back body curve, with a significant angle on the front lower body curve from the waist to the hip line (pronounced Q5).

The number and percentage of correctly classified test subjects in a particular body posture type is determined using the classification matrix (Table 11). The highest percentage of correctly classified test subjects, that is, complete prediction of 100%, was determined for lower body posture types LP2 and LP3, followed by LP1 with 96.77% correctly classified test subjects. Although the classification matrix showed different accuracy of predicting type LP1, a total of 99.02% correctly classified test subjects indicates a very high prediction possibility of identifying test subjects in belonging to the particular lower body posture type.

Table 11.

Classification matrix for lower posture types

Posture type	Correctly classified test subjects [%]	Correctly classified test subjects [N]
Posture type	Correctly classified test subjects [%]	LP2 p = 0.412	LP3 p = 0.284	LP1 p = 0.304
LP3	100.00	32	0	0
LP2	100.00	0	39	0
LP1	96.77	0	1	30
Total	99.02	32	40	30

Note: bold text represent statistical significance i.e. p value <0.05

Analysis of whole body posture can be achieved by combining the two proposed methods; firstly, we analyze upper body posture and then lower body posture. The classification matrix was created using the results of discriminant function strength and structure. The order of variables analysis and classification of posture types are shown in Figure 11. Classification of a tested sample showed the presence of all nine possible body posture combinations, confirming high variability of body posture characteristics (Figure 11).

In addition, we studied the correlation of posture indicators with main body circumferences and with body widths calculated from scanned data as the difference between the distance of the front and back anthropometric points from the posterior vertical plane measured on the same body girth line.

There is no strong correlation determined between posture indicators and main body circumferences, which indicates that there is no correlation between posture types and body size in general. However, body widths in the sagittal view showed some moderate correlations with posture indicators (Table 12). Correlation with body widths showed the influence of body shape, especially the shape of characteristic circumference cross-sections, on posture types. Body width on the chest line, defined as variable ΔCF-CB, mostly correlates with indicator Q5 defined as the front lower body angle, which supports the theory of body balance and correlation of the chest and hip zones. Correlation with indicators Q1 and Q6 is expected since those are the angles of the upper body curve from c7 to the chest line and from c7 of the scapula. Body width on the waist line, defined as variable ΔByF-WB, correlates with indicator Q3 responsible for lower body posture from the waist to the hip line and with indicators Q4 and Q5 representing the front body curve, which is expected since front body curvature is mostly dependable on the position of the front belly anthropometric point. Body width on the hip line, defined as variable ΔAF-HB, correlates with indicator Q4 representing the angle of the front upper curve.

Table 12.

Determined correlations between body widths in the sagittal view and posture indicators

	r	r ²	p
CF-CB
Q1	0.269	0.072	0.006
Q6	0.251	0.063	0.011
Q5	0.406	0.165	0.000
ByF-WB
Q3	−0.349	0.122	0.000
Q4	−0.419	0.176	0.000
Q5	0.419	0.175	0.000
AF-HB
Q4	−0.219	0.048	0.027

Indicators Q1 and Q4 are most significant as criteria of belonging to a particular upper body posture type, both correlating with main body widths in the sagittal view. Indicator Q5 is most significant, a criterion of belonging to a particular lower body posture type, shows the highest correlation with measurements of body width on the chest and waist line.

Conclusions

The basic purpose of the performed research was to develop a method for identification of upper and lower body posture types, based on posture indicators calculated using the standardized anthropometric measurement data obtained from a 3D body scanner, which will be valid, reliable, and objective. The selection of anthropometric measurements obtained on the 3D body model according to the standard ISO 7520 ensures reliability and objectiveness of the measuring procedure, and thus the posture indicators are calculated from these measurements.

The calculation of posture indicators from measurements obtained using the 3D scanner also enables a wider use range of the posture classification method, in view of the increasing use of scanning technology in various fields.

Body posture can be presented by characteristic front and back body curves with specific angles, which can be calculated based on standardized measurements and used as indicators for the classification of posture types. Use of angles as posture indicators instead of standard linear dimensions reduces the number of necessary indicators. From the aspect of clothing construction, classification of body posture types based on body front and back curve angles enables a direct link between ranges of angle values and block pattern modifications of a garment for a particular body posture type. Separate analysis of upper and lower body posture types enables clearer analysis of the influence of variables on a particular type and covers a wider range of posture variability.

The proposed research hypothesis can be accepted based on the cluster analysis results with a division in three upper and three lower body posture types, followed by discriminant analysis of posture indicators with the established relationship among the determined body postures. According to the defined posture indicators, the developed method differentiates three different types of upper posture and three types of lower body posture. Classification of the whole body posture can be achieved by combining the two proposed methods. This approach enables nine possible posture classifications, all of which are present and can be found in real life. Specific combinations correspond to particular posture type definitions that can be found in the literature. As shown in Figure 12, posture combination UP3 and LP2 corresponds to the definition of good posture and regular spine curvature, UP1/UP2 represent kyphotic spine curvature, UP3/LP1 represent lordotic curvature, and UP1/LP1 kypholordotic, all defined as irregular postures. Other combinations include a prominent back scapula area and front abdomen curve, as shown in combination UP2/LP2, which can be defined as good posture with a prominent scapula (Figure 12).

The obtained results and the developed method for body posture classification can be applied in the process of computer garment design according to individual anthropometric characteristics. Using this approach, the previously developed methods of the authors for computer-based garment construction can be improved by developing methods of garment construction for targeted body posture types.^24,27,28,31 The parameterization method enables adjustment of pattern segment dimensions based on defined mathematical regularities. In this way, obtained angles defined as posture indicators can be reversed into the body measurements of specific widths and heights from which they are calculated and used for pattern modification. Also, the proposed method for posture type classification will be further analyzed in terms of testing the defined posture indicators on the female population and implementing the characteristics of defined posture types to the design and construction of female made-to-measure clothing. Due to the high diversity of human body shapes and with respect to the increasing customer demand for garment fit and comfort, morphological classifications, such as sampling into body posture types, can greatly improve sizing systems and garment construction, focusing on customization of products for specific body types.

Footnotes

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Croatian Science Foundation under the project number 3011, “Application of mathematical modelling and intelligent algorithms in clothing construction.”

References

Canhadas Belli

Chaves

Siriani de Oliveira

et al.

Analysis of body posture in children with mild to moderate asthma. Eur J Pediatr 2009; 168: 1207–1216.

Dunk

Lalonde

Callaghan

. Implications for the use of postural analysis as a clinical diagnostic tool: reliability of quantifying upright standing spinal postures from photographic images. J Manip Physiol Ther 2005; 28: 386–392.

Fedorak

Ashworth

Marshall

et al.

Reliability of the visual assessment of cervical and lumbar lordosis: how good are we?

Spine 2003; 28: 1857–1859.

Watson

AWS

Mac Donncha

. A reliable technique for the assessment of posture: assessment criteria for aspects of posture. J Sport Med Phys Fit 2000; 40: 260–270.

Palmer

Epler

. Fundamentals of musculoskeletal assessment techniques, Philadelphia, PA: Lippincott Williams & Wilkins, 1998.

Garrison

Read

. Fitness for everybody, California, CA: Mayfield Publishing, 1980.

Walk

Lindsey

Corbin

. Concepts of physical fitness, New York: McGraw-Hill Companies, 2000.

Stefanovic D, et al. Učestalost telesnih deformacija i lošeg telesnog drzanja školske dece na području SR Srbije. In: 1st conference of school medicine doctors, Split-Trogir, Croatia, 1972, pp.351–357.

Auxter

Pyfer

Huettig

. Principal and methods of adapted physical education and recreation, New York: WCB McGraw-Hill, 1997.

10.

Lima

LCdeO

Barauna

Sologurem

MJJ

et al.

Postural alterations in children with mouth breathing assessed by computerized biophotogrammetry. J Appl Oral Sci 2004; 12: 232–237.

11.

Werghi N and Xiao Y. Recognition of human body posture from a cloud of 3D data points using wavelet transform coefficients. In: 5th IEEE international conference on automatic face and gesture recognition (ed A D Wiliams), Washington, DC, USA, 20–21 May 2002, pp.70–75. Washington, DC: IEEE. doi:10.1109/AFGR.2002.1004135.

12.

Treleaven

Wells

JCK

. 3D body scanning and healthcare applications. Computer 2007; 40: 28–34.

13.

Paušić

Cavala

Katic

. Relations of the morphological characteristic latent structure and body posture indicators in children aged seven to nine years. Coll Antropol 2006; 30: 621–627.

14.

Kosinac

. Kineziterapija sustava za kretanje, Split, HR: University of Split, 2002.

15.

Barrow

Mec Gi

. Measuring in physical education, Belgrade: Vuk Karadzic, 1975.

16.

Paušić J, Kosinac Z and Pažanin R. Evaluation of the reliability and validity of the scoliometer measuring instrument for the valorisation of the body posture indicators. In: international scientific and professional conference “sport-recreation-fitness”, Split, Croatia, 2005, pp.101–105. Split: University of Split Faculty of Natural Sciences and Education.

17.

Iunes

Bevilaqua-Grossi

Oliveira

et al.

Comparative analysis between visual and computerized photogrammetry postural assessment. Rev Bars Fizioter 2009; 13: 308–315.

18.

McEvoy

Grimmer

. Reliability of upright posture measurements in primary school children. BMC Musculoskel Dis 2005; 6: 35–35.

19.

Pausic J. The construction and the evaluation of new posture screening methods in boys aged 10 to 13. PhD Thesis, University of Split, Croatia, 2007.

20.

Bellemare

Males: understanding sizing requirements for male apparel. In: Faust

Carrier

(eds). Designing apparel for consumers, Philadelphia, PA: Woodhead Publishing Limited, 2014, pp. 189–220.

21.

Fisher T, et al. Automatic morphological classification with Case-Based Reasoning. In: 7nd international conference on 3D body scanning technologies (ed N D’Apuzzo), Lugano, Switzerland, 30 November–01 December 2016, paper no. 16.148, pp.148–158. Ascona: Hometrica Consulting.

22.

Surville JM. Posture, 3D real body, virtual try-on: toward fashion. In: 2nd international conference on 3d body scanning technologies (ed N D’Apuzzo), Lugano, Switzerland, 25–26 October 2011, paper no. 11.225, pp.225–233. Ascona: Hometrica Consulting.

23.

Huang

Mok

Kwok

et al.

Determination of 3D necklines from scanned human bodies. Text Res J 2011; 81: 746–756.

24.

Petrak

Mahnic

Rogale

. Impact of male body posture and shape on design and garment fit. Fibres Text East Eur 2015; 23: 150–158.

25.

Cho

Komatsu

Inui

et al.

Individual pattern making using computerized draping method for clothing. Text Res J 2006; 76: 646–654.

26.

Kim

Lee

Hong

et al.

Use of three-dimensional technology to construct ergonomic patterns for a well-fitting life jacket of heterogeneous thickness. Text Res J 2014; 85: 816–827.

27.

Petrak S. The method of 3D garment construction and cutting pattern transformation model. PhD Thesis, University of Zagreb, Croatia, 2007.

28.

Petrak

Rogale

. Systematic representation and application of a 3D computer-aided garment construction method, part I: 3D garment basic cut construction on a virtual body model. Int J Cloth Sci Tech 2006; 18: 179–187.

29.

Zhong

. Three-dimensional garment dressing simulation. Text Res J 2009; 79: 792–803.

30.

Kim

LeBat

. An exploratory study of users’ evaluations of the accuracy and fidelity of a three-dimensional garment simulation. Text Res J 2013; 83: 171–184.

31.

Mahnic

Petrak

. Investigation of the fit of computer-based parametric garment prototype. J Fiber Bioeng Informat 2013; 6: 51–61.

32.

Petrak S, Mahnic M and Ujevic D. Research of 3D body models computer adjustment based on anthropometric data determined by laser 3D scanner. In: international conference on 3D body scanning technologies (ed N D’Apuzzo), Lugano, Switzerland, 16–17 October 2012, paper no. 12.115, pp.115–126. Ascona: Hometrica Consulting.

33.

Chen

. Analysis of upper physical characteristics based on angle measurements. Text Res J 2011; 81: 301–310.

34.

ISO 20685:2010. 3D scanning methodologies for internationally compatible anthropometric databases.

35.

ISO 7250:1996. Basic body measurements for technological design.

36.

ISO 8559:1989. Garment construction and anthropometric surveys – body dimensions.

37.

Pavlić

. Statistical theory and application, Zagreb: Tehnička knjiga, 1970.

38.

Cooley

Lohnes

. Multivariate data analysis, New York: John Wiley & Sons Inc., 1971.

39.

Lachenbruch

. Discriminant analysis, New York: Hafner Press, 1975.