Scanner Benchmarking for SizeUK

Mr. R. M. Allen^(1,2), Ms. J Bougourd⁽³⁾,Mr. R A J Staples⁽¹⁾, Dr. C Orwin⁽⁴⁾,Dr. M Bradshaw⁽⁴⁾

(1) DLO DC RPS, Ministry of Defence, Skimmimgdish Lane Caversfield, Bicester, Oxon, OX27 8TS, UK

(2) Shape Analysis Ltd, 3 Brick Cottages, Brook Road, Aldham, Colchester, CO6 3RN, UK www.shapeanalysis.com

(3) London College of Fashion, 20 John Princes Street, London, UK

(4) Department of Textiles, DeMonfort University, Leicester, UK

ABSTRACT

This paper discusses the methodology used to assess 3D whole-body acquisition systems for use in the UK National Sizing Survey - SizeUK.

The objective of SizeUK was to conduct a sizing survey of approximately 5,000 men and 5,000 women in order to promote improvements in retail garment fit and sizing standards. The sizing survey was conducted using 3D whole-body acquisition systems, both to reduce the measurement time and cost per subject, and also enabling body shape to be captured, which is not possible by manual tape measurement.

Keywords: Scanning, Benchmarking, 3D, Whole, Body, Scanner, Sizing, Survey, Anthropometry.

1. Background

To enable the selection of the most suitable 3D body scanning systems, the project conducted a Scanner Bechmarking exercise of candidate 3D scanners. This consisted of 2 phases; Phase 1 – Scanner Manufacturers’ Self-Assessment, Phase 2 – 3D Centre Assessment.

Overall, the selected 3D scanner(s) were required to collect accurate body measurements in a rapid and cost-effective way from a large proportion of the UK population, including men, women and children. The survey subjects were to be scanned wearing underwear.

Guideline requirements for satisfactory scanning are set out below:

1. Size – a wide range of body sizes, e.g., for women US size 4 to 24, EU size 36 to 54, UK size 6 to 26.

2. Skin colour - from very light to very dark skin pigmentation.

3. Head hair - light colour to black; length short to long; loose or styled. A manufacturer may choose to specify that some subjects must wear head caps, or that some subjects’ heads must be measured manually.

4. Body hair - up to thick body hair. Again, the manufacturer must specify any necessary procedures in order to obtain accurate body surface measurements.

5. Perspiration - scanner to deliver accurate body surface data in the presence of normally encountered levels of perspiration.

The manufacturer needs to specify any necessary conditions or instructions for satisfactory operation of that scanner, including pose, breathing, etc. These conditions should also specify the underwear to be worn (e.g. fabric type, colour, style).

2. Scanner Benchmarking – Phase 1

Phase 1 was conducted entirely by the scanner manufacturers for convenience, and to reduce time and costs. Phase 1 is composed of four parts; (i), a questionnaire asking general questions about the scanner (Table 1); (ii), a form to record 6 measurements taken from a mannequin; (iii), a form for 9 measurements taken from a female subject (ideally this subject will be dark skinned); (iv), a declaration of which of 132 desired measurements the scanner can capture, both currently and with further short-term development, along with the corresponding measurement accuracy (where known). Short-term, for the purpose of these requirements, is defined as no longer than 2 months from the date of the declaration. The manufacturers were requested to provide an indication of the measurement accuracy of their scanner.

Table 1

Contact Details
		Product Name
		Contact Person
		Company Name
		Postal Address
		Web Address
		Telephone Number
		Fax Number
		Email Address
Body Measuring Systems
	Retail cost of system			$
	Rental cost of system			$
	What type of PC or workstation is required?
				Volts	Amps
	Power supply
			Length	Width	Height
	Dimensions of whole-body capture device
	Minimum floor space for a whole working system (excl. dressing room)
	Dimensions of the active scanning volume
3D Data
	Number of capture heads
	Number of sample points
	Time to capture a whole body
	Can users access the raw data points?
	Colour image (texture) captured of subject?
	Formats for data exchange
	Typical data file size
Scanning Pose
	Number of scans per subject required to collect data of whole body
	Scanning poses of subject (e.g. body erect; legs and arms apart)
Scanning Subjects
	Min. and max. size of women who can be successfully scanned and processed
	Min. and max. size of men who can be successfully scanned and processed
	Min. and max. size of children who can be successfully scanned and processed
Measurement Extraction
	How many body measurements can be extracted fully automatically?
	How many body measurements can be extracted semi-automatically?
	Are there facilities for manual measurement extraction?

The questionnaire aimed to gather basic information on the scanning system, such as size, scanning speed, hardware characteristics, software characteristics, subject pose, accuracy and cost.

Each scanner manufacturer was asked to carry out two scanning tasks, and a fill in a paper questionnaire these were;

A. Mannequin Scan

The first data gathering exercise was to scan and manually measure a clothing mannequin, and extract six measurements :

1. Bust – the maximum horizontal bust girth, as measured on the mannequin.

2. Waist – the minimum horizontal circumference on the mannequin.

3. Hips – the horizontal hip girth, as measured on the mannequin round the buttocks at the level of maximum circumference.

4. Bust height – the vertical distance between the horizontal bustline of the mannequin and the ground.

5. Waist height - the vertical distance between the horizontal waistline of the mannequin and the ground.

6. Hip height - the vertical distance between the horizontal hip level of the mannequin and the ground.

Table 2

Measurements extracted automatically or manually from scans
Total number of scans
	Measurement			Number of scan measurements within range of manual measurement
	Manual	Smallest from scan	Largest from scan		±0.5cm	±1cm	±1.5cm	>1.5cm
Bust girth
Waist girth
Hip girth
Height of bust from floor
Height of waist from floor
Height of hip from floor

B. Female Subject Scan

The second data gathering exercise is to scan and manually measure a female subject, ideally with a dark skin. The measurements to be gathered are based on the draft European Standard CEN/TC 248/WG 10 N 34 (Size Designation of Clothes – definitions and body measurement procedures):

1. Horizontal Dimensions

Bust girth – the maximum horizontal girth measured during normal breathing with the subject standing erect and the tape-measure passed horizontally under the armpits (axillae) and across the bust prominence.

Underbust girth – the horizontal girth of the body immediately below the breasts,

Waist girth – the girth of the natural waistline between the top of the hip bones (iliac chest) and the lower ribs, with the subject breathing normally and standing erect with the abdomen relaxed.

Hip girth – the horizontal girth measured round the buttocks at the level of maximum circumference.

2. Vertical Dimensions

Height – the vertical distance between the crown of the head and the soles of the feet, measured with the subject standing erect without shoes and with the feet together.

Bust height – the vertical distance between the bust level and the soles of the feet.

Waist height – the vertical distance between the waist level at the right side and the soles of the feet.

Hip height – the vertical distance between the hip level and the soles of the feet.

Inside leg length – the distance between the crotch and the soles of the feet, measured in a straight line with the subject erect, feet slightly apart, and the weight of the body equally distributed on both legs.

For the live scan, the subject was scanned a number of times (e.g. five). For each scan the subject entered the system, had a scan taken, and then left and re-entered the measurement system for the next scan. It had to be specified whether the measurements were extracted automatically or manually from the scans.

Table 3

Measurements extracted automatically or manually from scans
Total number of scans
	Measurement			Number of scan measurements within range of manual measurement
	Manual	Smallest from scan	Largest from scan	±1cm	±2cm	±4cm	>4cm
Bust girth
Underbust girth
Waist girth
Hip girth
Height
Bust height
Waist height
Hip height
Inside leg length

C. Measurement Declaration

Manufacturers were asked to tick which of the measurements their scanner was capable of recording, along with measurement accuracy where known (e.g. + 2cm). The full list of measurements is not contained in this paper.

3. Scanner Benchmarking – Phase 2

Based on the results of the Phase 1 Assessment, a subset of systems were proposed for the next round of tests listed in Table 3.

Table 4

Company / System	Address	Contact name	Nearest system	Automated measurement extraction
Cyberware WB4 and WBX	2110 Del Monte Avenue Monterey California 93940 USA	Steve Addleman Steve@cyberware.com	USA Natick, NIOSH	ü men only
Hamamatsu Photonics UK	2 Howard Court, 10 Tewin Road Welwyn Garden City Hertfordshire AL7 1BW England	Duncan Stacey Duncan.stacey@hamamatsu.co.uk Atsushi Tsujimura a.tsujimura@hamamatsu.co.uk 01707 294888	London / Germany / Boston	ü
Inspeck, Inc.	3530 St-Laurent Blvd. Suite 303, Montreal QU H2X 2V1, Canada	Christian Rochefort rochefort@inspeck.com +1 514 284 1101	Montreal?	û
[TC]² Textile/Clothing Technology Center	211 Gregson Drive Cary NC 27511-7909 USA	David Bruner Dbruner@tc2.com +1 919 380 2156	[TC]², USA	ü
TecMath Vitus Pro and Vitus Smart	Human Modeling Sauerweisen 2 Kaiserslautern Germany	Christian Lott Lott@hs.tecmath.com +49 (0)6301 606 300	Germany	ü
Telmat	6 Rue De L'Industrie 68360 Soultz France	Jean-Loup Rennesson	LCF, London, UK	ü
Wicks and Wilson Limited	Morse Road Basingstoke Hampshire RG22 6PQ	Bob Brash Bobb@wwl.co.uk 01256 842211	LCF, London, UK	Software from [TC]²

Phase 2 Methodology

Analysis of the systems can be divided into four categories:

(i) Visual appraisal of data quality:

Basic scan assessment

Skin colour, size & shape of subjects, degree of hairiness, Subject Movement

Fabric tests

(ii) System operation, etc.:

Scanner operation and usability

Subject acceptability

Software features and usability

System costs and service agreement

For each system tests will be carried out at the test site (on-site tests) over three days:

(iii)On-site questions and observations:

Subject Acceptability

Capturing, Storing… Retrieving Scans

Size and Shape Software features

Service Agreement

Measurement extraction (can be done after on-site testing)

(iv) Measurement extraction from scans of mannequin and living subjects:

Manual measurement extraction from scans, using interactive tools

Automated measurement extraction using proprietary software

Automated measurement extraction using standard software (Hammamatsu)

4. Analysis of Results

Visual appraisal

To ensure image appears to have minimal data loss due to any of the factors above e.g. movement and measurement volume. Fabric samples were used to assess the scanners performance against various coloured, contrasting (either within a pattern or against skin tone) or surface finished fabrics.

System Operation

To assess ease of operation, i.e. skill level of operator and extent of automation.

On-site Questions and observations

All systems were found suitable by all subjects. There were no problems in data storage or retrieval. Each manufacturer demonstrated their systems capabilities with regards to size and shape software. There were great variations between systems on price, rental agreements etc.

Measurement Extraction

The main criteria for choosing a scanner for the SizeUK survey was measurement repeatability and number of measurements that were statistically, significantly the same as the manual measurements taken during the benchmarking.

The statistics test used was the t-test. However some of the scanning systems were so variable on their measurement extraction that limits had to be applied to the measurement variation. These values were taken from Chapter VII of ‘1988 Anthropometric Survey of US Army Personnel: Methods and summary statistics – Gordon et al’.

The following method was used:-

• 4 Teams of Measurers

• Each Subject was Landmarked in morning and measured twice by one set of measurers.

• Each subject was re-measured by a new team using the same landmarks from the first session.

– Landmarks were only refreshed if required.

• Process repeated on day 2 but teams changed roles.

• The mean of the absolute differences was calculated for each pair of investigators, for each dimension.

• The Allowed Error was the rounded maximum of the four measures.

Table 5 Allowable Errors From US Army Survey

• Where the Allowed Error is matched in the U.S Army Survey their Allowed Error is used.

• Where no matched measurement exists, then 1.0 cm is assumed.

The two following graphs illustrate the reasons for using allowable errors:-

Graph 1 Graph 2

Scannner with small variation between scans Scanner with large variation between scans

The significance of the above example is that a scanner with very large variations between scans can ‘statistically’ perform very well in the t-test. But obviously they are so variable that they would perform extremely badly in use. If however applied limits of variation are set then these scanning systems will be eliminated and only systems displaying a high degree of reproducibility will pass.

During the evaluation process it was found that several systems exhibited large variations in results between scans, this could be due to several reasons:-

1. Errors in the raw scan data. This can be caused by the presence of outliers or holes in the scan causing spurious measurements.
2. Poor landmark detection. The inability to locate body landmarks efficiently and consistently.
3. Poor scan capture. Incomplete or missing data.
4. Poor or simple segmentation of body parts

In summary some scanning systems showed extremely variable results, which displayed no trend or offset that could be determined, these errors made it impossible to recommend the use of these scanners for SizeUK. However 2 scanning systems displayed repeatability levels greater than manual measurement and were therefore judged as suitable for SizeUK.

However the final factor that swung the decision was the fact that the [TC]² software was customisable to such an extent, that only 9 out of the 157 measurements required for SizeUK needed to be extracted manually. These were:-

Weight, Height, Back Neck to Elbow (bent), Back Neck to Wrist (bent), Hand Length, Hand Girth, Head Girth, Head loop , Armscye Girth.

5. Conclusion

SizeUK has developed a procedure that can be used successfully to assess the performance of 3D scanning systems for use in anthropometric surveys. The system that was chosen displayed all the characteristics of the original goals i.e.

Overall, the selected 3D scanner(s) are required to collect accurate body measurements in a rapid and cost-effective way from a large proportion of the UK population, including men, women and children. The survey subjects will be scanned wearing underwear.

The exception was for children. No scanning system was acceptable for children due to movement during capture, or poor data capture, we await future developments to solve this problem.