Sage Journals: Discover world-class research

Abstract

Research purpose: This study aimed to lay a research foundation for smart hand wearable design by classifying the right-hand data of 4545 adults aged 20 to 69. Further, to increase the practical applicability of the hand classification system, a hand type discrimination method and regression equations for the hand dimensions of each type were presented. Methods: This study statistically analyzed eighth Size Korea data with IBM SPSS Ver.26.0. Cluster analysis was performed to classify both finger length and circumference type. Discriminant analysis was conducted, yielding discriminant functions to aid potential smart hand wearable wearers in self-diagnosing their hand types. Linear regression analysis yielded regression equations for the detailed finger dimensions for the pattern-making of smart hand wearables. Results: The finger length type was categorized into four types: the Uphill type, Downhill type, Mountain type, and Horizon type for both men’s and women’s hands. The finger circumference type was categorized into two types, the Cone and Cylinder types, for both men’s and women’s hands. The discriminant function showed a mean accuracy rate of 89.9% and the regression equations a mean explanatory power of 72.9%. Conclusion: The hand classification system proposed in this study aimed to improve the fingertip fit of smart hand wearable products by analyzing the configuration of motion tracking gloves or haptic gloves. In addition, considering the practical applicability for both wearers and designers of smart hand wearables, a discrimination method of finger types for wearers’ self-diagnosis and regression functions of finger dimensions for designers’ pattern making were provided.

Keywords

Hand classification smart gloves smart hand wearables glove fit finger type 3D scanning

Introduction

Smart wearable technology has dramatically advanced the interaction between reality and the virtual world through extended reality (XR) systems,¹ among which controller-type smart devices operated by hand contribute greatly to increased immersion in the virtual world by providing haptic feedback to users. However, hand-held smart devices have limited naturalness because users cannot grasp or touch objects in a virtual environment while holding the controller.² Thus, the development of real-world immersive XRs requires different kinds of devices that facilitate relatively more natural human interactions, particularly by freeing users’ hands, recognizing gestures, and providing tactile feedback to help users feel virtual experiences as real ones.³ For this reason, various types of smart hand wearables in the form of gloves are being developed, allowing users to touch and manipulate virtual objects in a more intuitive and direct way.

Given the intricate structure of the hand, attaining an optimal fit for gloves presents a considerable challenge. Notwithstanding these difficulties, it is crucial to achieve a precise fit by conducting a thorough analysis of hand ergonomics and anthropometrics because glove fit significantly impacts hand safety, functionality, and comfort, all of which influences the decision to wear gloves.⁴ A tight-fitting glove can restrict the wearer’s hand movement, impede circulation in the fingers, and increase the risk of work-related injuries.^5,6 Conversely, if the glove is too large or loose-fitting, the wearer may experience reduced dexterity and grip capability.^5,6

While the use of elastic materials is one of the most effective ways to address fitting issues in gloves, the same approach does not apply to smart hand wearables. Smart hand wearables often feature a structure involving strings or rigid links attached to the fingers, known as exoskeleton types,⁷ that lack flexibility. Even with the utilization of elastic fabrics, the inclusion of sensors, actuators, wires, and other components^8,9 can significantly reduce flexibility. Therefore, the fit of smart hand wearables cannot heavily rely on material elasticity, and fit degradation issues cannot be completely eliminated regardless of the materials used. Consequently, enhancing the fit of smart hand wearables should be based on a thorough examination of the wearer’s hand morphology.

Smart hand wearables are worn primarily for hand motion tracking and haptic feedback.^10–12 These functions can be implemented accurately when the fit in the fingertip is excellent. For example, smart hand wearables that track finger motions detect movement through sensors at the fingertips, and those that provide haptic feedback provide tactile stimulation by attaching vibration motors such as linear reactive actuators (LRAs) to the fingertips.¹³ Thus, to provide accurate tracking of hand movements and realistic tactile feedback, the hand wearable’s fit of all five fingertips must be precise.

Studies measuring and analyzing hand dimensions have been conducted among people in several countries, including Bangladesh,^14,15 Czech Republic,¹⁶ Filipin,¹⁷ India,^18–21 Jordan,^22,23 Nigeria,^24,25 the USA, the Netherlands, and Italy.²⁶ In Korea, several studies^27–30 classified hand shapes based upon statistical analyses of hand dimension data retrieved from Size Korea, the National Standard Anthropometric Surveys led by the Korean Agency for Technology and Standards, Government of Korea, or hand data collected by researchers themselves, discussed below.³¹

Choi and Kim³⁰ classified the hands of 824 adults aged 18 to 64 into two types: short and chubby and long and slender. Kwon and Choi²⁹ classified the hands of 134 men and 131 women aged 18 to 30 into two male types—small, short-fingered and large, and long-fingered, and three female types—medium-sized, thick-fingered, and thin-fingered. Kim and Kee²⁸ classified the hands of 325 men aged 20 and above into two types: thin and long and thick and blunt. Choi and Do³¹ measured the right hands of 353 older women aged 60 and above and classified them into three types: average, long and wide, and short and narrow. These four studies successfully classified the overall characteristics of hands by applying factor analysis to statistically determine the criteria items for hand classification. However, solely focusing on overall characteristics may overlook the intricate details of the fingers. Jee and Yun²⁷ classified the hands of 321 adults into four types: spacious hand and short fingers, short palm but above-average fingers, long palm and fingers, and narrow hand and short fingers, and emphasized that, in addition to breadth and length of hands, other dimensions such as finger length should be considered for improved glove fitting.

Moreover, relative to these studies, Vergara et al.³² developed a classification system more closely related to the characteristics of smart hand wearables. They analyzed the metacarpal lengths and the length from the wrist point to the fingertips of 139 Spanish adults. This analysis was significant as it considered the variance in hand shape that occurs due to each of the five fingers. However, achieving an accurate fit for the smart hand wearables' finger region relies more on the individual lengths of each finger rather than the overall length from the wrist to the fingertips. Therefore, to develop a classification system that better suits smart hand wearables, it is crucial to directly estimate the lengths of individual fingers. Additionally, shape classification based on the circumference of the fingers, not addressed in these studies, should also be considered.

Therefore, to address this gap, this study develops, for the first time, a classification system for a smart-hand wearable design that incorporates information on individual finger length and circumference by extracting hand measurements that elucidate finger characteristics. Another novelty of the study is that to enhance the practical applicability of the hand classification system, it proposes guidelines for its convenient implementation, including hand-type diagnosis and pattern making.

Materials and methods

Participants

This study analyzed the right-hand data of 4545 adults aged 20 to 69 (2020 men and 2525 women) who participated in the three-dimensional scan measurements conducted during the eighth Size Korea project in 2020 and 2021. After excluding missing values, the finalized data of 2018 men and 2524 women were classified. The distribution by age group of the participants is shown in Table 1.

Table 1.

Distribution of participants by age group.

		20s	30s	40s	50s	60s	Total
Men	Frequency	847	495	297	152	227	2018
Men	Percentage (%)	42.0	24.5	14.7	7.5	11.2	100.0
Women	Frequency	868	690	470	264	232	2524
Women	Percentage (%)	34.4	27.3	18.6	10.5	9.2	100.0

Data collection

In eighth Size Korea, nine hand items were measured through direct measurement and 39 hand items through three-dimensional scan measurement. The length, first joint circumference, and end joint circumference of the five fingers used in this study were collected through three-dimensional scanning (Figure 1).

Figure 1.

Hand dimensions and symbols.

Data analysis

Classification of finger length and circumference types

Analysis of the hand length type was performed for all items, using an index value based on the middle finger length to compare the differences between finger lengths. To this end, hierarchical cluster analysis was conducted by setting the values of thumb length ( $ℓ_{t}$ ), index finger length ( $ℓ_{i}$ ), ring finger length ( $ℓ_{r}$ ), and pinky length ( $ℓ_{p}$ ), divided by middle finger length ( $ℓ_{m}$ ), as independent variables. The square Euclidean distance was used as a measure of similarity, and categorized by Ward’s method. After analyzing the dendrogram extracted through hierarchical cluster analysis to determine the final cluster level and the number of clusters, K-means cluster analysis was performed. Data categorized by cluster analysis were post-tested with the ANOVA Duncan test to confirm that each independent variable showed a significant difference for each cluster. Finally, 4545 three-dimensional hand-scan figures were analyzed to extract representative shapes corresponding to each type.

In the analysis of the hand circumference type, the index value of the end joint circumference ( $e c$ )/first joint circumference ( $f c$ ) of each finger was used to compare the shape according to the circumference of the fingers. Therefore, the $e c$ / $f c$ of each of the five fingers was set as an independent variable. Hierarchical cluster analysis, K-means cluster analysis, and representative figure extraction were performed in the same way as for finger length type. In addition, t-tests were performed for independent samples to confirm that each independent variable showed a significant difference for each cluster. Finally, a cross-analysis of finger length types and circumference types was performed on the data distribution of type combinations.

Discrimination analysis for finger type diagnosis

Discriminant analysis was applied to determine the finger length and circumference types. First, stepwise discriminant analysis was conducted to identify the independent variables important in determining finger length and circumference types, and three or four items were selected according to importance level to derive the discriminant function and accuracy rate. The discriminant function was calculated in the form of Fisher’s linear discriminant.

Regression analysis for pattern making

Linear regression analysis was conducted to provide regression equations of detailed finger dimensions. Considering practical applications, the independent variables input into the regression equation were limited to one for the length type and two for the circumference type, and the explanatory power and validity of the derived regression equation were analyzed.

All statistical analyses were performed using IBM SPSS Ver. 26.0.

Results and discussion

Classification by finger length type

The results of the hierarchical cluster analysis using the ratio of the other finger’s lengths based on the middle finger length show that both men and women can be categorized into four clusters. K-means cluster analysis was used to classify Korean men’s and women’s hands into Uphill, Downhill, Mountain, and Horizon types, as shown in Table 2.

Table 2.

Finger length-type classification of men’s and women’s hands.

Finger length type	Uphill	Downhill	Mountain	Horizon	—
Finger length type					—
Men (n = 2018)
	n = 668 (33.1%)	n = 450 (22.3%)	n = 480 (23.8%)	n = 420 (20.8%)	—
Representative 3D scan data					F
Mean $ℓ_{t}$ / $ℓ_{m}$	0.64 C	0.67 B	0.60 D	0.71 A	1703.937***
Mean $ℓ_{i}$ / $ℓ_{m}$	0.90 B	0.90 AB	0.88 C	0.90 A	35.485***
Mean $ℓ_{r}$ / $ℓ_{m}$	0.96 A	0.91 B	0.90 C	0.94 A	248.824***
Mean $ℓ_{p}$ / $ℓ_{m}$	0.73 B	0.67 D	0.67 C	0.76 A	904.082***
Women (n = 2524)
	n = 786 (31.1%)	n = 617 (24.4%)	n = 471 (18.7%)	n = 650 (25.8%)	—
Representative 3D scan data					F
Mean $ℓ_{t}$ / $ℓ_{m}$	0.63 B	0.70 A	0.63 B	0.70 A	1130.135***
Mean $ℓ_{i}$ / $ℓ_{m}$	0.90 C	0.91 B	0.89 D	0.92 A	99.141***
Mean $ℓ_{r}$ / $ℓ_{m}$	0.92 B	0.92 C	0.90 D	0.95 A	342.925***
Mean $ℓ_{p}$ / $ℓ_{m}$	0.72 B	0.69 C	0.64 D	0.76 A	1830.970***

***p < .001, Duncan test, A > B.

In the Mountain type, as the middle finger is generally longer than all other fingers, the difference between the five fingers is remarkable, while in the Horizon type, all other fingers are proportionately longer than the middle finger, reflecting a very small difference in length between all five fingers. Conversely, in the Uphill type, the thumb and the index finger are proportionately shorter than the middle finger, while the ring and little fingers are proportionately longer. Finally, the Downhill type has a proportionately longer thumb and an index finger than the middle, while the ring and little fingers are shorter. The post-test for the cluster analysis found that all four independent variables showed a significant difference between types at the 0.001 level.

The classification results highlight that human hands not only exhibit size variations resulting from size reduction and enlargement while maintaining the same shape and proportion but also vary significantly based on the lengths of the five fingers. This observation has been supported by previous studies. For instance, Vergara et al.³² conducted an analysis of the variability in the lengths of metacarpals and fingers, proposing nine combinations of three palm shapes (PS) and three hand shapes (HS). Among these combinations, the PS1-HS1 type, characterized by a thumb-index predominant pattern, shows similarities to the Downhill type in our study. Conversely, the PS3-HS3 type, characterized by the ring-little predominant pattern, shows similarities to the Uphill type in our study. Furthermore, the overall average PS2-HS2 type encompasses both the Mountain and Horizon types observed in our study.

The findings show the similarities between the hand types identified by Vergara et al. (2019)³² and the finger length types in our study, although there are noticeable differences in the frequencies of each type. PS1-HS1, PS2-HS2, and PS3-HS3 types account for 19.4%, 22.3%, and 11.5% of the total participants, respectively. Notably, the least common is the PS3-HS3 type, which is similar to the Uphill type, but the latter exhibits the highest proportion among all types in our study, with 33.1% for men and 31.1% for women. These differences in frequencies could be attributed to variations in classification criteria between Vergara et al.³² and our study, as well as potential distinctions between the Spanish and Korean people.

Classification by finger circumference type

According to the results of the hierarchical cluster analysis using the ratio of the first joint circumference based on the end joint circumference of each finger, both men and women can be categorized into two clusters. Thus, K-means cluster analysis was used to classify men’s and women’s hands into the Cone and Cylinder types, as shown in Table 3.

Table 3.

Finger circumference type classification of men’s and women’s hands.

Finger circumference type	Cone	Cylinder	—	—
Finger circumference type			—	—
Men (n = 2018)
	n = 1012 (50.1%)	n = 1006 (49.9%)	—	—
Representative 3D scan data			t	p
Mean ${e c}_{t}$ / ${f c}_{t}$	0.89	0.94	−36.050	0.000***
Mean ${e c}_{i}$ / ${f c}_{i}$	0.73	0.76	−28.875	0.000***
Mean ${e c}_{m}$ / ${f c}_{m}$	0.76	0.80	−27.511	0.000***
Mean ${e c}_{r}$ / ${f c}_{r}$	0.81	0.83	−15.659	0.000***
Mean ${e c}_{p}$ / ${f c}_{p}$	0.83	0.86	−17.502	0.000***
Women (n = 2524)
	n = 1284 (50.9%)	n = 1240 (49.1%)	—	—
Representative 3D scan data			t	p
Mean ${e c}_{t}$ / ${f c}_{t}$	0.84	0.90	43.405	0.000***
Mean ${e c}_{i}$ / ${f c}_{i}$	0.73	0.77	33.219	0.000***
Mean ${e c}_{m}$ / ${f c}_{m}$	0.76	0.80	29.923	0.000***
Mean ${e c}_{r}$ / ${f c}_{r}$	0.81	0.83	12.506	0.000***
Mean ${e c}_{p}$ / ${f c}_{p}$	0.83	0.85	14.648	0.000***

***p < .001.

In the Cone type, the circumference difference between the first and end joints of each finger is large, reflecting that the shape of the finger narrows steeply toward the end. Conversely, the Cylinder type shows a relatively small circumferential difference between the first and end joints of each finger, such that the shape of the finger is fairly cylindrical. In addition, the independent samples’ t-test of the cluster analysis found that five independent variables showed a significant difference between types at the 0.001 level.

As evident from the second classification results, the diversity of finger lengths in human hands is not the only distinguishing factor; a clear distinction can be made based on the circumference of the fingers. However, no study has discussed the detailed morphological diversity based on the specific circumference of the fingers. Kwon and Choi²⁹ divided women’s hands divided into three types: medium-sized, thick-fingered, and thin-fingered and focused the discussion on the circumference of the fingers. Unfortunately, they categorized fingers as thick or thin based on overall finger thickness without distinguishing between the root and the tip of the fingers.

This study argues that taking into account the difference in circumference between the root and the tip of the fingers is crucial for the accurate functioning of smart hand wearables. For example, a wearer with noticeable differences in circumference between the finger root and fingertip, such as a cone type, would wear gloves that fit the finger root. However, this can result in excess material at the fingertip area of the gloves, potentially causing poor adherence to attached motion sensors or tactile actuators and leading to malfunctions. By contrast, a wearer with similar circumferences between the finger root and the fingertip, such as a cylinder type, may feel that the gloves are tight around the blunt fingertips when wearing gloves according to the overall hand size. This is especially important for smart hand wearables as they often position sensors or actuators at the fingertips,¹³ resulting in a loss of material flexibility. In the worst-case scenario, the fingertips may not fit into the wearable device at all. Therefore, to avoid fitting issues faced by wearers of each type, it is necessary to design glove patterns for each of the hand classification types proposed in this study and manufacture smart hand wearables accordingly.

Cross-analysis of finger length and circumference types

A cross-analysis of men’s finger length and circumference types found that the combination of the Horizon and Cone types accounted for the lowest proportion at 9.4%, whereas the combination of the Uphill and Cone types had the highest proportion at 17.6%. The appearance rates of all type combinations were evenly distributed with a difference of less than 10% points (Table 4). The corresponding cross-analysis for women found that the combination of the Mountain and Cylinder types accounted for the lowest proportion at 7.8%, whereas that of the Uphill and Cylinder types showed the highest proportion at 17.1%. The data for women were also found to be evenly distributed among types, with a difference of less than 10% points in the incidence of all type combinations (Table 5).

Table 4.

Cross-analysis of the finger length and circumference types (Men).

		Length types				Total
		Uphill	Downhill	Mountain	Horizon	Total
Circumference types	Cone	356 (17.6%)	207 (10.3%)	260 (12.9%)	189 (9.4%)	1012 (50.1%)
Circumference types	Cylinder	312 (15.5%)	243 (12.0%)	220 (10.9%)	231 (11.4%)	1006 (49.9%)
Total		668 (33.1%)	450 (22.3%)	480 (23.8%)	420 (20.8%)	2018 (100.0%)

The combination with the highest frequency and the combination with the lowest frequency among all finger type combinations are highlighted in bold.

Table 5.

Cross-analysis of the finger length and circumference types (Women).

		Length types				Total
		Uphill	Downhill	Mountain	Horizon	Total
Circumference types	Cone	354 (14.0%)	286 (11.3%)	273 (10.8%)	371 (14.7%)	1284 (50.9%)
Circumference types	Cylinder	432 (17.1%)	331 (13.1%)	198 (7.8%)	279 (11.1%)	1240 (49.1%)
Total		786 (31.1%)	617 (24.4%)	471 (18.7%)	650 (25.8%)	2524 (100.0%)

The combination with the highest frequency and the combination with the lowest frequency among all finger type combinations are highlighted in bold.

Discrimination analysis for finger type diagnosis

While the successful classification of the existing hand data is important, the practical application of the proposed classification system in this study relies on the availability of a discriminant method for new data. Therefore, we introduced a discriminant function that enables the diagnosis of hand types, ensuring its practical applicability.

Stepwise discriminant analysis was used to find the differences between men and women in the items selected as important. For example, thumb length and little finger length were chosen as the most critical items for finger length type discrimination for men and women, respectively (Table 6). Moreover, the joint circumferences of the index finger and the middle finger were deemed important for finger length type discrimination for men and women, respectively (Table 7).

Table 6.

Accuracy rate of the finger length type discriminant function according to the number of items.

Number of items	Men		Women
Number of items	Added items	Accuracy rate, (%)	Added items	Accuracy rate, (%)
1	$ℓ_{t}$	37.1	$ℓ_{p}$	37.7
2	$ℓ_{m}$	64.5	$ℓ_{m}$	64.9
3	$ℓ_{p}$	90.2	$ℓ_{t}$	89.8
4	$ℓ_{r}$	94.5	$ℓ_{r}$	94.1
5	$ℓ_{i}$	96.2	$ℓ_{i}$	97.5

The ultimately selected number of items is highlighted in bold.

Table 7.

The accuracy rate of the finger circumference type discriminant function according to the number of items.

Number of items	Men		Women
Number of items	Added items	Accuracy rate, (%)	Added items	Accuracy rate, (%)
1	${f c}_{t}$	62.5	${f c}_{t}$	65.1
2	${e c}_{t}$	77.8	${e c}_{t}$	80.5
3	${f c}_{i}$	82.9	${f c}_{m}$	85.3
4	${e c}_{i}$	88.0	${e c}_{m}$	91.6
5	${f c}_{m}$	89.2	${f c}_{i}$	91.8
6	${e c}_{m}$	91.1	${e c}_{i}$	94.6
7	${f c}_{r}$	92.5	${e c}_{p}$	95.5
8	${e c}_{r}$	94.4	${f c}_{p}$	97.4
9	${e c}_{p}$	95.3	${e c}_{r}$	97.5
10	${f c}_{p}$	98.6	${f c}_{r}$	98.4

The ultimately selected number of items is highlighted in bold.

We found that the more measurement items included in the classification function, the higher the accuracy of type determination. However, measurement items cannot be added endlessly, given the time and effort required to measure them. By comparing the accuracy rate of the finger length type discriminant function by the number of items, a severe loss of accuracy (approximately 25%) was found between three items and two items for both men’s and women’s hands (Table 6). Moreover, the accuracy rate dropped considerably for four versus fewer items in the finger circumference type discriminant function (Table 7). Thus, the measurement items were limited to three for length type discrimination and four for circumference type discrimination in the interest of accuracy rate and practical ease of use.

As a result, it was deemed appropriate to use the three items of the length of the thumb, middle finger, and little finger for both men and women for finger length type discrimination. For finger circumference type discrimination, it was deemed appropriate to use the first and end joint circumference of the thumb and index fingers for men but the first and end joint circumference of the thumb and middle fingers for women.

The discriminant function was calculated in the form of Fisher’s linear discriminant by reflecting the hand-dimension items selected through stepwise discriminant analysis (Tables 8 and 9). The accuracy rate was 90.2% in finger length type discrimination for men, 89.8% in finger length type discrimination for women, 88.0% in finger circumference type discrimination for men, and 91.6% in finger circumference type discrimination for women. Group scatterplots of finger length types according to the canonical discriminant functions one and two are shown in Figure 2. As group scatterplots are applicable when there are three or more groups, group scatterplots for finger circumference types, which are classified into two categories, were not generated in this study.

Table 8.

Discriminant function of finger length type.

	Length type	Discriminant function	Accuracy rate, (%)
Men	Uphill	0.711* $ℓ_{t}$ + 1.892* $ℓ_{m}$ + 0.898* $ℓ_{p}$ – 118.052	90.2
	Downhill	1.316* $ℓ_{t}$ + 2.218* $ℓ_{m}$ – 0.095* $ℓ_{p}$ – 120.006
	Mountain	−0.206* $ℓ_{t}$ + 3.192* $ℓ_{m}$ – 0.049* $ℓ_{p}$ – 123.424
	Horizon	2.212* $ℓ_{t}$ + 0.755* $ℓ_{m}$ + 1.105* $ℓ_{p}$ – 120.192
Women	Uphill	0.099* $ℓ_{p}$ + 2.688* $ℓ_{m}$ + 0.801* $ℓ_{t}$ – 120.574	89.8
	Downhill	−1.297* $ℓ_{p}$ + 3.647* $ℓ_{m}$ + 0.865* $ℓ_{t}$ – 124.079
	Mountain	0.918* $ℓ_{p}$ + 1.365* $ℓ_{m}$ + 1.824* $ℓ_{t}$ – 117.571
	Horizon	−0.388* $ℓ_{p}$ + 2.303* $ℓ_{m}$ + 1.902* $ℓ_{t}$ – 120.524

Table 9.

Discriminant function of finger circumference type.

	Circum type	Discriminant function	Accuracy rate, (%)
Men	Cone	1.180* ${f c}_{t}$ + 4.129* ${e c}_{t}$ – 0.306* ${f c}_{i}$ + 0.879* ${e c}_{i}$ – 191.934	88.0
Men	Cylinder	1.838* ${f c}_{t}$ + 3.366* ${e c}_{t}$ + 0.338* ${f c}_{i}$ + 0.000* ${e c}_{i}$ – 188.320	88.0
Women	Cone	0.447* ${f c}_{t}$ + 2.930* ${e c}_{t}$ – 0.095* ${f c}_{m}$ + 2.602* ${e c}_{m}$ – 181.063	91.6
Women	Cylinder	1.337* ${f c}_{t}$ + 3.851* ${e c}_{t}$ – 0.802* ${f c}_{m}$ + 1.810* ${e c}_{m}$ – 180.824	91.6

Figure 2.

Group scatterplot of finger length types by canonical discriminant functions one and two (Left: Men, Right: Women).

Regression analysis for pattern making

When designing a smart hand wearable pattern, it would be ideal to have all the dimensions of the wearer’s hands. However, since such customization is practically impossible for off-the-shelf products, this study attempted to develop a regression equation based on a limited set of hand dimensions so that patterns can be designed with only a few given dimensions. Regression analysis was performed by assuming that the dimensions of only one finger were given in turn, and the final independent variable was determined by comparing the explanatory power of all regression equations. Accordingly, the independent variable selected is the middle finger length for the length type regression equation and the first and end joint circumference of the middle finger for the circumference type regression equation.

The analysis of the results of the regression for the thumb, index finger, ring finger, and pinky length against the length of the middle finger (Table 10) shows all regression equations significant at the 0.001 level, and all the values of the variance inflation factor (VIF) at less than 10, indicating no problem of multicollinearity. According to the

R^{2}

value, the explanatory power of the finger length type regression equation ranged from at least 51.2%–86.6%, with a mean of 75.6%.

Table 10.

Regression equation by finger length type.

	Type	Item	Regression equation	$R^{2}$	F
Men	Uphill	$ℓ_{t}$	0.617* $ℓ_{m}$ + 1.675	0.747	1968.088***
		$ℓ_{i}$	0.901* $ℓ_{m}$ - 0.291	0.795	2585.287***
		$ℓ_{r}$	0.846* $ℓ_{m}$ + 6.901	0.839	3476.708***
		$ℓ_{p}$	0.687* $ℓ_{m}$ + 3.555	0.767	2196.089***
	Downhill	$ℓ_{t}$	0.591* $ℓ_{m}$ + 5.974	0.734	1239.255***
		$ℓ_{i}$	0.911* $ℓ_{m}$ - 0.908	0.797	1752.000***
		$ℓ_{r}$	0.847* $ℓ_{m}$ + 4.711	0.836	2278.756***
		$ℓ_{p}$	0.685* $ℓ_{m}$ – 1.200	0.703	1059.772***
	Mountain	$ℓ_{t}$	0.576* $ℓ_{m}$ + 1.750	0.730	1292.396***
		$ℓ_{i}$	0.882* $ℓ_{m}$ + 0.107	0.822	2205.994***
		$ℓ_{r}$	0.849* $ℓ_{m}$ + 4.348	0.866	3094.051***
		$ℓ_{p}$	0.674* $ℓ_{m}$ + 0.009	0.707	1158.829***
	Horizon	$ℓ_{t}$	0.657* $ℓ_{m}$ + 3.974	0.718	1064.050***
		$ℓ_{i}$	0.861* $ℓ_{m}$ + 3.084	0.751	1264.635***
		$ℓ_{r}$	0.854* $ℓ_{m}$ + 6.201	0.811	1795.899***
		$ℓ_{p}$	0.657* $ℓ_{m}$ + 6.696	0.649	777.965***
Women	Uphill	$ℓ_{t}$	0.519* $ℓ_{m}$ + 7.927	0.687	1722.851***
		$ℓ_{i}$	0.867* $ℓ_{m}$ + 2.638	0.808	3306.957***
		$ℓ_{r}$	0.844* $ℓ_{m}$ + 5.702	0.840	4119.192***
		$ℓ_{p}$	0.702* $ℓ_{m}$ + 1.097	0.794	3015.947***
	Downhill	$ℓ_{t}$	0.591* $ℓ_{m}$ + 5.974	0.512	644.673***
		$ℓ_{i}$	0.886* $ℓ_{m}$ + 0.416	0.784	2230.031***
		$ℓ_{r}$	0.824* $ℓ_{m}$ + 5.317	0.808	2583.507***
		$ℓ_{p}$	0.697* $ℓ_{m}$ – 3.829	0.682	1321.551***
	Mountain	$ℓ_{t}$	0.554* $ℓ_{m}$ + 9.945	0.565	608.836***
		$ℓ_{i}$	0.873* $ℓ_{m}$ + 3.504	0.776	1625.420***
		$ℓ_{r}$	0.841* $ℓ_{m}$ + 7.443	0.811	2006.009***
		$ℓ_{p}$	0.729* $ℓ_{m}$ + 2.370	0.774	1609.270***
	Horizon	$ℓ_{t}$	0.610* $ℓ_{m}$ + 6.175	0.699	1506.394***
		$ℓ_{i}$	0.896* $ℓ_{m}$ + 0.885	0.780	2295.270***
		$ℓ_{r}$	0.877* $ℓ_{m}$ + 2.895	0.833	3227.467***
		$ℓ_{p}$	0.772* $ℓ_{m}$ – 2.086	0.757	2017.430***

***p < .001.

Next, the regression analysis results of the first and end joint circumference of the thumb, index finger, ring finger, and little finger against those of the middle finger (Table 11) show that all regression equations are significant at the 0.001 level. In addition, the values of the variance inflation factor (VIF) are all less than 10, indicating no multicollinearity problem. According to the

R^{2}

value, the explanatory power of the finger circumference type regression equation ranged from 45.2% to 86.7%, with a mean of 70.1%.

Table 11.

Regression equation by finger circumference type.

	Type	Item	Regression equation	$R^{2}$	F
Men	Cone	${f c}_{t}$	0.083* ${f c}_{m}$ + 0.768* ${e c}_{m}$ + 23.532	0.512	528.450***
		${e c}_{t}$	0.197* ${f c}_{m}$ + 0.576* ${e c}_{m}$ + 21.996	0.641	900.852***
		${f c}_{i}$	0.683* ${f c}_{m}$ + 0.259* ${e c}_{m}$ + 8.196	0.829	2447.645***
		${e c}_{i}$	0.100* ${f c}_{m}$ + 0.720* ${e c}_{m}$ + 6.483	0.822	2322.790***
		${f c}_{r}$	0.281* ${f c}_{m}$ + 0.553* ${e c}_{m}$ + 11.946	0.693	1138.188***
		${e c}_{r}$	0.173* ${f c}_{m}$ + 0.646* ${e c}_{m}$ + 4.151	0.749	1503.102***
		${f c}_{p}$	0.166* ${f c}_{m}$ + 0.535* ${e c}_{m}$ + 11.015	0.594	737.835***
		${e c}_{p}$	0.194* ${f c}_{m}$ + 0.509* ${e c}_{m}$ + 3.176	0.672	1037.361***
	Cylinder	${f c}_{t}$	0.017* ${f c}_{m}$ + 0.858* ${e c}_{m}$ + 26.712	0.474	452.408***
		${e c}_{t}$	0.177* ${f c}_{m}$ + 0.611* ${e c}_{m}$ + 20.255	0.692	1127.743***
		${f c}_{i}$	0.672* ${f c}_{m}$ + 0.259* ${e c}_{m}$ + 9.467	0.862	3121.344***
		${e c}_{i}$	0.110* ${f c}_{m}$ + 0.680* ${e c}_{m}$ + 7.308	0.836	2548.797***
		${f c}_{r}$	0.257* ${f c}_{m}$ + 0.593* ${e c}_{m}$ + 11.806	0.750	1504.590***
		${e c}_{r}$	0.149* ${f c}_{m}$ + 0.653* ${e c}_{m}$ + 4.774	0.777	1745.489***
		${f c}_{p}$	0.186* ${f c}_{m}$ + 0.563* ${e c}_{m}$ + 8.607	0.666	1000.822***
		${e c}_{p}$	0.189* ${f c}_{m}$ + 0.480* ${e c}_{m}$ + 4.106	0.723	1311.197***
Women	Cone	${f c}_{t}$	−0.078* ${f c}_{m}$ + 1.021* ${e c}_{m}$ + 25.360	0.452	527.377***
		${e c}_{t}$	0.186* ${f c}_{m}$ + 0.650* ${e c}_{m}$ + 15.609	0.727	1709.608***
		${f c}_{i}$	0.635* ${f c}_{m}$ + 0.343* ${e c}_{m}$ + 6.929	0.867	4169.434***
		${e c}_{i}$	0.090* ${f c}_{m}$ + 0.704* ${e c}_{m}$ + 6.937	0.802	2593.505***
		${f c}_{r}$	0.217* ${f c}_{m}$ + 0.661* ${e c}_{m}$ + 9.382	0.724	1678.913***
		${e c}_{r}$	0.114* ${f c}_{m}$ + 0.694* ${e c}_{m}$ + 4.151	0.748	1900.747***
		${f c}_{p}$	0.233* ${f c}_{m}$ + 0.473* ${e c}_{m}$ + 8.971	0.654	1211.720***
		${e c}_{p}$	0.157* ${f c}_{m}$ + 0.459* ${e c}_{m}$ + 6.679	0.621	1048.251***
	Cylinder	${f c}_{t}$	−0.045* ${f c}_{m}$ + 0.876* ${e c}_{m}$ + 26.031	0.468	543.907***
		${e c}_{t}$	0.182* ${f c}_{m}$ + 0.672* ${e c}_{m}$ + 15.686	0.692	1388.372***
		${f c}_{i}$	0.631* ${f c}_{m}$ + 0.323* ${e c}_{m}$ + 7.453	0.844	3333.820***
		${e c}_{i}$	0.129* ${f c}_{m}$ + 0.716* ${e c}_{m}$ + 4.608	0.805	2556.400***
		${f c}_{r}$	0.281* ${f c}_{m}$ + 0.573* ${e c}_{m}$ + 9.519	0.708	1497.822***
		${e c}_{r}$	0.151* ${f c}_{m}$ + 0.654* ${e c}_{m}$ + 4.194	0.746	1817.051***
		${f c}_{p}$	0.209* ${f c}_{m}$ + 0.524* ${e c}_{m}$ + 7.921	0.652	1156.737***
		${e c}_{p}$	0.177* ${f c}_{m}$ + 0.467* ${e c}_{m}$ + 5.787	0.626	1037.012***

***p < .001.

The hand classification system proposed in this study has enhanced the generality of finger length-based classification by yielding similar results that reinforce previous research in this area. Furthermore, it has revealed variations in proportion distribution based on race or nationality, even among similar types. Moreover, we have expanded the analysis of detailed finger morphology by introducing finger circumference types, which have not been explored in extant research. The cross-analysis of finger length and circumference types has consistently demonstrated a balanced frequency distribution across all combinations, highlighting the meaningfulness of all proposed types in this study. Additionally, by deriving and providing discriminant functions and regression equations based on our classification system, we anticipate an increased practical utility.

Conclusion

With recent advances in smart hand wearables, such as motion-tracking gloves and haptic gloves for XR, there is a growing demand for detailed finger shape analysis by incorporating various sensors or motors at the fingertips. The categorization of fingers is necessary because human hands do not simply scale up or down in size while maintaining the same proportions of finger length and circumference. Our hands exhibit variations not only in overall size but also in the individual ratios of finger lengths and circumferences. Unfortunately, this crucial aspect has been consistently overlooked, both in research and in the actual production of gloves, resulting in poor fit and inadequate performance. To innovatively address this issue, this study proposes a new hand classification system focusing on the length and circumference characteristics of the fingers for designing smart hand wearable devices.

The academic significance of this study lies in the classification of hand types based on the detailed characteristics of the finger area rather than general aspects of the entire hand. The cross-analysis results indicated an even distribution of frequency among all length and circumference type combinations without excessive bias. Therefore, the hand classification system proposed in this study successfully categorizes the Korean hand into eight distinctive types and would be very useful if applied to pattern-making or sizing systems of smart hand wearables in the future.

The practical implication of the proposed finger classification system is its potential to improve fingertip fit, thereby enhancing the effectiveness of smart hand wearable products such as motion-tracking gloves or haptic gloves. Additionally, considering the practical applicability for both wearers and designers of smart hand wearables, this study provides a discrimination method for finger types to enable hand type diagnosis and regression equations for finger dimensions to assist pattern-making.

The study has a significant limitation in that the data only comprise Korean samples. Future research could extend this to other nationalities. Moreover, future research could build on the proposed classification method and validate its effectiveness through real-life assessments of wearability to enhance the robustness and practical relevance of this study’s findings.

Footnotes

Declaration of conflicting interests

The author(s) 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 New Faculty Startup Fund grant funded by the Seoul National University (No. 350-20220075) and the National Research Foundation of Korea (NRF) grant funded by the Korean Government (MSIT) (No. 2022R1A4A5034046).

Ethical approval

This study was approved by Seoul National University Institutional Review Board (Ethics Code: E2211/002–006) on 14 November 2022.

ORCID iDs

Yujin Hong

Hye Suk Kim

Hee Eun Choi

References

Ozioko

Dahiya

. Smart tactile gloves for haptic interaction, communication, and rehabilitation. Adv Intell Syst 2022; 4: 2100091.

Bowman

McMahan

Ragan

. Questioning naturalism in 3D user interfaces. Commun ACM 2012; 55: 78–88.

Johnson-Glenberg

. Immersive VR and education: embodied design principles that include gesture and hand controls. Front Robot AI 2018; 5: 81.

Griffin

Kim

Carufel

, et al. Dimensions of the dynamic hand: implications for glove design, fit, and sizing. Advances in interdisciplinary practice in industrial design. Cham: Springer International Publishing, 2019, pp. 38–48, In: Proceedings of the AHFE 2018 international conference on interdisciplinary practice in industrial design, 21–25 July 2018.

Torrens

Campbell

Tutton

. Design issues in military footwear and handwear. In: Advances in military textiles and personal equipment. Cambridge, UK: Woodhead Publishing, 2012, pp. 139–164.

Hsiao

Whitestone

Kau

, et al. Firefighter hand anthropometry and structural glove sizing: a new perspective. Hum Factors 2015; 57(8): 1359–1377.

Kim

Park

Jung

. Effect of' RAPAEL smart glove’s on cognitive function and activities of daily living in mild cognitive impairment. Soc Occup Ther Aged Dement 2018; 12(2): 75–85.

Balzano

Minieri

Stranieri

. Mandri: a new proposal of manus VR facility integration in everyday car driving. Web, artificial intelligence and network applications. Cham: Springer International Publishing, 2019, pp. 860–869, In: Proceedings of the workshops of the 33rd international conference on advanced information networking and applications (WAINA-2019) 33.

Bein

Gofman

. Designing a unity game using the haptic feedback gloves, VMG 30 plus. Advances in Intelligent Systems and Computing. Cham: Springer International Publishing, 2020, vol. 1134, pp. 393–400, In: 17th international conference on information technology–new generations (ITNG 2020).

10.

Glauser

Panozzo

, et al. Interactive hand pose estimation using a stretch-sensing soft glove. ACM Trans Graph 2019; 38: 1–15.

11.

Jörg

Mueller

, et al. Virtual hands in VR: motion capture, synthesis, and perception. In: SIGGRAPH Asia 2020 courses, Virtual event, 4–13 December 2020, pp. 1–32.

12.

Wang

Song

Naqash

, et al. Toward whole-hand kinesthetic feedback: a survey of force feedback gloves. IEEE Trans Haptics 2019; 12: 189–204.

13.

Caeiro-Rodríguez

Otero-González

Mikic-Fonte

, et al. A systematic review of commercial smart gloves: current status and applications. Sensors 2021; 21: 2667.

14.

Imrhan

Sarder

Mandahawi

. Hand anthropometry in Bangladeshis living in America and comparisons with other populations. Ergonomics 2009; 52: 987–998.

15.

Khadem

Islam

. Development of anthropometric data for Bangladeshi male population. Int J Ind Ergon 2014; 44(3): 407–412.

16.

Bures

Gorner

Sediva

. Hand anthropometry of Czech population. In: Rachman

Ratnayake

RMC

(eds) IEEE international conference on industrial engineering and engineering management (IEEM), Singapore, 6–9 December 2015, pp. 1077–1082, New York, NY: IEEE.

17.

Del Prado-Lu

. Anthropometric measurement of Filipino manufacturing workers. Int J Ind Ergon 2007; 37(6): 497–503.

18.

Nag

Desai

. Hand anthropometry of Indian women. Ind J Med Res 2003; 117: 260–269.

19.

Dewangan

Owary

Datta

. Anthropometric data of female farm workers from north eastern India and design of hand tools of the hilly region. Int J Ind Ergon 2008; 38(1): 90–100.

20.

Meena

Dangayach

Bhardwaj

. Measuring anthropometric data for designing hand tools in handicraft industries. Int J Process Manag Benchmarking 2013; 3(3): 334–351.

21.

Dewangan

Patel

Vidhu

et al. An investigation of the hand anthropometric database of agricultural workers and integration of the database into tools and protective gear designs. Work 2023; 74(4): 1461–1480.

22.

Mandahawi

Imrhan

Al-Shobaki

, et al. Hand anthropometry survey for the Jordanian population. Int J Ind Ergon 2008; 38: 966–976.

23.

Chandra

Chandna

Deswal

. Analysis of hand anthropometric dimensions of male industrial workers of Haryana state. Int J Engg 2011; 5(3): 242–256.

24.

Okunribido

. A survey of hand anthropometry of female rural farm workers in Ibadan, Western Nigeria. Ergonomics 2000; 43(2): 282–292.

25.

Ismaila

. Anthropometric data of hand, foot and ear of university students in Nigeria. Leonardo J Sci 2009; 15(8): 15–20.

26.

Robinette

Daanen

Paquet

, et al. The CAESAR project: a 3-D surface anthropometry survey. In: Second international conference on 3-D digital imaging and modeling (cat. No. PR00062), Ottawa, ON, 08–08 October 1999, pp. 380–386, IEEE.

27.

Jee

Yun

. An anthropometric survey of Korean hand and hand shape types. Int J Ind Ergon 2016; 53: 10–18.

28.

Kim

Kee

. Classification and identification of Korean hand shapes based on anthropometric hand data analysis. J Korea Saf Manag Sci 2012; 14: 75–85.

29.

Kwon

Choi

. A study on classification of hand shape. J Korean Soc Costume 2005; 55: 11–19.

30.

Choi

Kim

. The database development of 2-D and 3-D hands measurement for improving fitness of gloves-focused on the classification of hand type and analysis of 3-D hand shape. J Korean Soc Cloth Text 2004; 28: 1300–1311.

31.

Choi

. Analysis on hand types of elderly women. Fash Text Res J 2013; 15: 574–582.

32.

Vergara

Agost

Bayarri

. Anthropometric characterisation of palm and finger shapes to complement current glove-sizing systems. Int J Ind Ergon 2019; 74: 102836.

Development of a hand classification system for smart hand wearables

Abstract

Keywords

Introduction

Materials and methods

Participants

Data collection

Data analysis

Classification of finger length and circumference types

Discrimination analysis for finger type diagnosis

Regression analysis for pattern making

Results and discussion

Classification by finger length type

Classification by finger circumference type

Cross-analysis of finger length and circumference types

Discrimination analysis for finger type diagnosis

Regression analysis for pattern making

Conclusion

Footnotes

Declaration of conflicting interests

Funding

Ethical approval

ORCID iDs

References