Human Body Dimensions Extraction From 3D Scan

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

2010 International Conference on Intelligent Computation Technology and Automation

Human Body Dimensions Extraction From 3D Scan Data

Xiaozhi Li 1, 2 Xiaojiu Li 1, 2
1. School of Art and Clothing
2. Key Laboratory for Advanced Textile Composite of Ministry of Education
Tianjin Polytechnic University, TJPU
Tianjin, P. R. China
E-mail: shelly_lixz@yahoo.cn
E-mail: lixiaojiu02@yahoo.com.cn

Abstract—3D body dimensions extracting is the basis that 3D


Body dimensions extracting
body scanner evolve in clothing applications. In this paper, by 3D body scanner
the body point cloud is acquired by 3D body scanner
developed by our team. The initial point cloud is processed
for reconstructed body triangular mesh model which is used Apparel CAD
to extract the body dimensions. The methods of processing system
the point cloud and constructed body model are simply
discussed and the detail discussion can be found in my paper
that is being published. In order to calculate the body Clothing styles
dimensions, the reference points are identified according to selecting
the characteristic of the feature contours. Then the body
dimensions, including length size and girth size, are
extracted. By measuring 10 feature sizes of different bodies, Automatically Modifying patterns
creating patterns interactively
the factors that affect the dimensional accuracy is analysed.
The result shows the method is effective to extracting the
body dimensions.
Single layer
cutting
Keywords-3D body scanner, body dimensions, made-to-
measure, 3D body model, reference points, feature dimensions
Quick sewing
1. INTRODUCTION system
With the development of 3D scanning technology, the Figure 1. Made-to-measure chain
applications of 3D body scanners are now rapidly
evolving into many areas, typical including 2. THE RECONSTRUCTION OF 3D BODY MODEL
anthropometric research, the clothing industry, the film
industry, computer animation and medicine [1] [2]. For In this paper, the 3D body scanner developed by our
clothing applications, the body is scanned by 3D body team [4] [5] is used to collect the body data. In the
scanner and the dimensions related to clothing are derived process of scanning the subject, in order to pick up the
from the digital body model. These dimensions are then data in the armpits and the inner thighs, upper limbs and
transferred to apparel CAD systems for designing clothes. torso has the angle about 30~45 degree and the lower
The patterns are sent to cut bed to cut single layer. This is limbs separate naturally with two legs flushing with the
so-called made-to-measure (MTM) which creates the shoulder. After the whole body is scanned, scan line point
individualized clothes. The specific process is shown as cloud [6] of the rear and the front of the body are
Fig.1. reconstructed (shown as Fig.3(a), the points of arms are
MTM is compatible with the current trend that people not got) and the points of every scan line are stored in
pursue and has been the main means of clothes producing computer in sequence. Because of physical properties of
[3]. The 3D body scanning and body dimensions measuring device, environment of measurement, light and
extracting are the basis of MTM as Fig.1 describes. the defects of calibration algorithm, the initial body point
cloud include some noisy points and holes. In order to
improve the surface reconstruction speed and quality, the
initial point cloud should be processed shown as Fig.2.

978-0-7695-4077-1/10 $26.00 © 2010 IEEE 441


DOI 10.1109/ICICTA.2010.849
Initial scan line point cloud

The front and the rear registering

Point cloud denoising and smoothing

Filling holes

Point cloud reducing

Figure 2. The initial point cloud processing

By registering the corresponding markers pasted on


the body before scanning the body, the front and the rear
can be integrated as a whole body model [7]. The noisy
points can be removed by the distance-based algorithm Figure 4. Body triangular mesh model
and the method of mean curvature flows [5]. Then the
points of every scan line are fitted into a closed curve by 3. 3D BODY DIMENSIONS EXTRACTING
the method of cubic parametric spline and the holes A. Identification of reference points
disappear. The points are so dense along the measuring
direction that they are not suitable for reconstructing the The human body dimensions mainly include body
surface model. The algorithm of chord length between height, neck girth, breast girth, waist girth, hip girth and
consecutive points to reduce the point data [8] is crotch height. The additional dimensions include shoulder
employed to reduce the point cloud. The processed point width, waist height, and hip height and so on. The basic
cloud is shown as Fig.3 (b). idea of extracting body dimensions is that the body is
divided into several parts where is used to search the
feature position in proportion of plus and minus 5% range
[11]. In order to get the dimensions, the reference point
should be found out at first.
The armpit points and crotch point have been
identified when the body triangular mesh model is
constructed. The neck point includes front neck point
(FNP), side neck point (SNP) and back neck point (BNP)
[12] [13]. The SNP is the hollowest point of the part that
connects the head and torso of human orthographic
projection. The digital model of the body standing front,
find the x-value of best side point of every scan line, if the
y point satisfies the equation as following:
x
z m _ xi1  m _ xi ! 0&&m _ xi  m _ xi 1  0 (1)

(a) Initial point cloud (b) Processed point cloud Then this point is SNP shown as Fig.5. The root of the
neck can be determined by SNP, so the FNP and BNP are
Figure 3. 3D body point cloud easy to identify.
The processed point cloud is divided into four parts:
human torso above underarm points, human torso below
underarm points, left leg and right leg. Every part is
constructed into polygon mesh by the method of one-to-
one branching [9] and this method is also used to connect
the human torso above underarm points and the human
Figure 5. Front contour of the body
torso below underarm points. The connection between
human torso and legs refers to the method of one-to-many Breast point (BP) is the highest point of the body
branching [10]. The body triangular mesh model is breast. In order to determine the waist line, the exact
constructed shown as Fig.4. location of the waist line must be ascertained by the
method of identify a waist point (WP). According to the

442
proportion of body, the waist point is the most concave In which y p is the y-coordinate of the cross section,
1
point at the waist of the body profile. Hip point (HP) is
the highest point of the body hip. The digital model of the D is the ratio coefficient, ( xn , yn , zn ) and ( xm , ym , zm ) are
body standing side, BP, WP and HP can be identified by the coordinates of the intersection edge’s endpoints.
the method of identifying the SNP shown as Fig.6. Choose a point on cross section and find the nearest edge
from it as the first intersection edge. Then find the
triangle that the intersection edge belongs to and the other
edge that intersects with this triangle is the second
intersection edge whose node with cross section is the
second point of girth line. Find other triangle intersecting
with the second edge and identify the third intersection
edge. Go on this method until all intersection points are
found out. Connect the intersection points by order and
the girth line is constructed which is made up by many
small segments. The sum of the small segments’ length is
the size of girth line. By this method, the breast girth line,
waist girth line and hip girth line can be extracted by the
point of BP, WP and HP shown as Fig.7.

Figure 6. The body profile

Shoulder points (SP) are the endpoints of the body


shoulder that connect upper limbs and torso. When the
digital model of the body stands front, the x-value’s
difference of the most side points of adjacent scan line is (a) Bustline (b) Hipline
larger at shoulders than at arms. According to this
characteristic, find a threshold G through, if the point
satisfies the equation as following:

° m _ xi 1  m _ xi  G
­ (c) Neck contour (d) Waistline
® (2)
°̄ m _ xi 1  m _ xi ! G Figure 7. Cross-section contour

Then it is shoulder point shown as Fig.5. C. Result of body dimensions


Using VC++ and OpenGL to program the above
B. Extraction of feature dimensions of the body algorithm, the body model is reconstructed and the
After the reference points identified, it is easy to dimensions can be extracted automatically. Table 1 is the
calculate the feature dimensions of the body. There are body dimensions of Fig.4.
two types of body dimensions: one is the length size, the
other is the girth size. The length size is easy to calculate, TABLE 1. TABLE 1 BODY DIMENSIONS IN FIG.4
such as body height, it is the vertical distance between the Unit:cm
highest point and lowest point of the body data. Intersect body height 163.5 waist height 103.7
a plane with the body triangular mesh model, and the hip height 78.3 crotch height 73.5
intersecting line is girth line. Then the girth size can be breast height 117.5 hip girth 87.8
calculated. The values of the points on the intersecting waist girth 64.0 breast girth 81.1
line can be calculated by pro rata relationship: neck girth 34.5 shoulder width 36.9

y p1  ym 4. ERROR ANALYSIS
D
yn  ym There are two factors that affect the dimensional
accuracy, one is the process of acquiring the body point
x p1 D ( xn  xm )  xm (3) cloud and the other is the process of extracting the body
dimensions.
z p1 D ( zn  zm )  zm In this paper, the 3D body scanner is developed by our
team and it takes 12s to scan a whole body. In the process
of scanning, it is difficult to ensure that the body is static
and the body is moving slightly in fact. So this will defect

443
the following measurement. When extracting the size of a [1] P.R.M. Jones, M. Rioux, P.R.M. Jones and M. Rioux, “Three-
certain part, the body posture should be standard, such as dimensional surface anthropometry: Applications to the human
body,” Optics and Lasers in Engineering, vol. 28, 1997, pp. 89–
identifying the armpit points, if the angle between arms 117, doi: 10.1016/S0143-8166(97)00006-7.
and torso is too small to make the upper arms together [2] H. A.M. Daanen and G. Jeroen van de Water, “Whole body
with torso, which results in armpit points lower. scanners,” Displays, vol. 19, Mar.1998, pp. 111–120, doi:
In the process of extracting the body dimensions, the 10.1016/S0141-9382(98)00034-1.
body cross-section contour lines should be continuous and [3] M. Xia, “Research on the Made to Measure System of the Men’s
smooth curves, the sub straight lines instead of smooth Individual Pattern Based on MTM,” Shanghai: Dong Hua
University,2006.
curves make the dimension smaller.
[4] Y. X. Wang and X. J. Li, “A portable body measurement system
5. CONCLUSION based on stereo vision”, Magic World of Textiles, 2008, pp. 686-
691.
The body point cloud is got by the 3D body scanner [5] X. Z. Li, X. J. Li, “3D body point cloud data denoising and
developed by our team and pre-processed for registration,” Proc of the 2nd International Conference on
reconstructing the body triangular mesh model which is Intelligent Computing Technology and Automation, 2009, pp.587-
590, doi: 10.1109/ICICTA.2009.376.
used to extract the body dimensions. A total of 6 feature
[6] L. g. Liu, “Studies on denoising and smoothing of point-based
points and 10 feature sizes are calculated. It takes 12s to model,” Hangzhou: Zhejiang University, 2007.
scanning the body, 3s to processing the point data and [7] X. Z. Li, “Human body data processing and 3D model
constructing the body triangular mesh model and 1s to reconstructing based on the technology of 3D non-contact human
extracting the body dimensions. All these time can be able body measurement, ” Tianjin: Tianjin Polytechnic University,
2007.
to satisfy extracting the body dimensions automatically.
[8] D. F. Rogers, J. A. Adams. “Mathematical elements for computer
Using this method to measuring a number of human graphics,” New York: McGraw-Hill, 1990.
bodies, the error of results is limited in 1.9cm compared [9] S. W. Chae, G. M. Lee, “Volume Triangulation from plannar cross
with the results by manual measurement which shows that sections,” Computer & Structures, vol. 72, July.1999, pp. 93-108,
this method is effective. doi:10.1016/S0045-7949(99)00026-7.
[10] J. Y. Zheng, “Acquring 3-D Models from Sequences of Contours,”
IEEE Trans. PAMI, vol. 16, Feb.1994, pp. 163-178, doi:
ACKNOWLEDGMENT 10.1109/34.273734.
[11] R. Pargas, “Automating Information Extraction From 3D Scan
The author would like to thank the help of Ph.D. Yu- Data,” Clemson University, 1998.
Xiu Wang and she also thanks the colleagues from [12] C. C. L. Wang, T. K. K. Chang , M. M. F. Yuen, “From Laser-
Department of clothing design and engineering and key scanned Data to Feature Human Model a System Based on Fuzzy
laboratory for advanced textile composite of ministry of Logic Concept,” Computer Aided Design, 35(2003), pp. 241 – 253,
education, Tianjin Polytechnic University. doi: 10.1016/S0010-4485(01)00209-3.
[13] Y. L. Ding. Ergonomics, Beijing: Beijing Institute of Technology
REFERENCES Press, 2005

444

You might also like