Precision of the CAESAR scan-extracted measurements

Abstract

Three-dimensional (3D) body scanners are increasingly used to derive 1D body dimensions from 3D whole body scans for instance, as input for clothing grading systems to make made-to-measure clothing or for width and depth dimensions of a seated workstation. In this study, the precision of the scanner-derived 1D dimensions from the CAESAR survey, a multinational anthropometric survey, was investigated. Two combinations of scanning teams with 3D whole body scanners were compared, one called the US Team and the other the Dutch Team. Twenty subjects were measured three times by one scanner and one team, and three times by the other combination. The subjects were marked prior to scanning using small dots, and the linear distances between the dots were calculated after processing the scans. The mean absolute difference (MAD) of the repetitions was calculated and this was compared to reported acceptable errors in manual measurements from the US Army's ANSUR survey when similar measurements were available. In addition, the coefficient of variation (CV) was calculated for all measurements. The results indicate that the CAESAR scan-extracted measurements are highly reproducible; for most measures the MAD is less than 5 mm. In addition, more than 93% of the MAD values for CAESAR are significantly smaller than the ANSUR survey acceptable errors. Therefore, it is concluded that the type of scan-extracted measures used in CAESAR are as good as or better than comparable manual measurements. Scan-extracted measurements that do not use markers or are not straight-line distances are not represented here and additional studies would be needed to verify their precision.

Introduction

Although three-dimensional (3D) anthropometric scanners have a vast potential to capture the shape of human bodies, they are increasingly used to determine 1D body dimensions like stature and circumferences. 1D body dimensions like chest, waist or hip circumferences have been used for centuries and, as a result, there are many databases available such as the data collations by Jürgens et al. (1990), Coblentz et al. (1992) and Churchill et al. (1977). Over the centuries, designers have developed techniques for using 1D measurements, whereas they may not have developed methods to use the 3D information.

Coward et al. (1997) demonstrated that reference points which are recognizable from the surface contours can be reliably extracted by human observers from scan images. However, 1D dimensions often use palpation of the bony parts underneath the skin in order to identify reference locations. The waist circumference, for instance, should be measured as the circumference halfway between the tenth rib (bottom-most palpable rib) and the iliocristale (top of the iliac crest of the pelvis on the side) points, according to several standards. In most cases, such as in this example, these points cannot be determined by merely examining the surface, although some exploratory work has been performed in this area (Li et al., 2003; Suikerbuik et al., 2004). As a result, unless markers placed over palpated points can be effectively identified in the 3D scan, the resulting measurements can be very inaccurate.

In the Civilian American and European surface anthropometry resource (CAESAR) 3D anthropometric survey (Robinette et al., 2002; Blackwell et al., 2002; Daanen and Robinette, 2001), 72 landmarks were palpated and marked with stickers prior to scanning so that they could be visualized and extracted from the scan images. Then, 1D measurements consisting of distances between the landmarks or between the landmarks and standing or seated surfaces were calculated. Since there is no consensus yet on how to calculate circumferences from 3D scans unambiguously, all circumferences were taken in the traditional manner in CAESAR, rather than being extracted from scans. Therefore, circumferences are not examined in this study. This paper examines the relative precision of this method of extracting 1D measurements as compared against published precision measurements of traditional 1D measurements from the ANSUR study (Gordon et al., 1989).

ANSUR uses the term mean absolute differences (MAD) to indicate the error between repeated measures of the same subjects. These values were used to derive a maximum allowable error for measurement in that survey. Team members were then trained and monitored to stay within this level of error. The purpose of this study was to identify if these allowed differences for manual measurement were realistic and achievable with the 3D scan method.