STAGE 3 : Sizing system
6.6 Automatic landmarking and data extraction software
The process of generating body dimensions automatically is distinct from the process of creating a 3-D digital body. Most scanner systems have the ability to automatically derive linear measurements from the scan, but the algorithms used by different scan-ner companies vary. The scanscan-ner that gescan-nerates the best scan for a project may not have the best automatic measuring system for that project. Most scanners can accept a scan in a generic format made by a different scanner, so it is possible to scan in one system and derive measurements in another.
The first step in generating a measurement from a scan is to identify the points on the body to be measured, the body landmarks. These landmarks are well defined by anthropometrists (Hotzman et al., 2011), tailors (Roetzel and Fritz, 2014), and apparel practitioners (Joseph-Armstrong et al., 2010), though each group has different sets of landmarks, different placement of landmarks, and different methods of measuring between landmarks. Because there is great variation among body shapes, it can be very difficult to place landmarks reliably, and although a trained anthropometrist will generally place landmarks more reliably, some variation will always be present (Kouchi and Mochimaru, 2011). It is even more difficult to design software to reliably place landmarks on body scans. This is generally done by programming a search for certain geometric features on the body (Suikerbuik et al., 2004), but programming for every possible body configuration can be almost impossible, particularly given the variation of shape between underweight and overweight bodies.
Because clothing is designed to move with the body, many of the critical landmarks that define measurements are where the body articulates—the joints. Therefore land-marks are often placed by anthropometrists and apparel practitioners by palpating to feel the appropriate bony protuberances that define the center of joint movement (Fig. 6.10). It is not possible for the scan software to achieve the same level of pre-cision and reliability by judging landmarks on joints based on the surface geometry of the body.
Other landmarks are established by common practice in the apparel industry or dic-tated by clothing styles. Placement of the landmark to measure waist circumference is a good example of this (Wren et al., 2014). The “natural” waist can be identified as the level of the spine of greatest articulation, between L4 and L5 of the lumbar vertebrae, or at the omphalion (belly button), or in normal weight women at the point where the body curves in the most at the side. In some body shapes, this area where the body curves in and makes a natural ledge is parallel to the floor, but in others, it tips down in front. On the other hand, most men have a small (or when less athletic, a substantial) bulge at the level of greatest articulation. However, none of this is helpful as currently
most clothing is worn lower than the natural waist and has been since the 1950s. One way to identify the waist, depending on the goals of the study, is to have each indi-vidual being scanned identify their preferred waist position for themselves, by placing a snug elastic band at their own waist. Most scan software provides a variety of choices for waist placement, so another choice is to measure the waist at several different locations.
Of course the scanner software can quite easily identify the smallest circumference on the torso between the high hip and the tenth rib—but this is generally a point even higher than any preferred landmark for the natural waist. In a similar issue the Fig. 6.10 Image showing skeletal landmarks, located by anthropometrists and apparel practitioners by palpating for the bony protuberances. Reliably finding the point on the bone closest to the center of the joint rotation requires training for some of the landmark locations.
landmark for the hip measurement is traditionally at the fullest protrusion of the but-tocks; a circumference is taken parallel to the floor at this point. However, the largest circumference of the lower body is generally lower, where the bulge of the thighs adds to the circumference of the body (Fig. 6.11).
Another example of landmarking that is only determined by apparel practice is in the placement of the side seam, the line down the side of the body that divides the front of the body from the back of the body. This is an important landmark for apparel pat-ternmaking because it is the dividing point to create arc measurements at the bust, hips, and waist that define the balance of the body. Most apparel practitioners can identify where the side seam should be placed to provide the best balance on different body types—but it is difficult to define this in terms that can be automatically located on a body scan reliably (Ashdown et al., 2008;Brownbridge et al., 2013).
Another area of the body that is complex and varied and therefore presents a problem for automated measurement extraction is the neckline measurement taken at the base of the neck (Huang et al., 2011). Though the cervicale at the back of the neck and the top of the sternum at the front of the neck are generally (but not always) easily identifiable on a scan, the actual placement of the neckline as it crosses the trapezius muscle on the side of the neck can be very difficult to determine. The precise curve of the neckline is also an apparel patternmaking construct. This measurement is therefore difficult when taken manually as well. Some apparel practitioners have devised a method for determining the desired curve using a chain that can be laid around the neck to determine the mea-surement more precisely than is possible with a tape measure.
Fig. 6.11 Representations of the traditional hip measurement taken at the greatest protrusion of the buttocks (the higher measurement) and the largest circumference value that is generally lower on the body (where the bulge of the thigh adds to the circumference). The end-use requirements of the study will determine which measurement is most useful.
Most scanner software developers recognize that the computer algorithms will not always place landmarks reliably and therefore make provision for easily indicating a different landmark point and retaking measurements. A trained operator can often improve the reliability and validity of a study by checking the measurements on a scan as they are taken and visually correcting landmark placement as necessary. In any case, scans should always be checked before the scanee leaves the booth to retake a scan if the person was not in the right position or if the scan is not optimized for some reason.
For optimal reliability of a study in which measurements will be derived, it is nec-essary to place landmarks manually, palpating for joint centers and marking the land-marks that are critical and cannot be reliably located automatically. If the scanner has capability to capture color or grayscale information along with the scan, then land-marks can be set with a washable marker or with an adhesive dot. If the scanner only takesXYZ coordinate data and no color information, a dimensional landmark can be used. Human Solutions makes a very precise dome, 20 mm in diameter that can be captured in the scan (seeFig. 6.12). As the size of this dome is known, the precise landmark point can be identified at the top of the dome and transferred down to the surface of the scan. The bump of the landmark can then be removed from the scan.
Of course the advantages of scanning over manual measurements are the compar-atively short amount of time needed with each study participant and the fact that highly trained assistants are not needed to conduct the study, as is true of a manually conducted anthropometric study. Manually placing landmarks adds time and requires skillful assistants, increasing the time and expense of an anthropometric study con-ducted with the scanner. However, if the number of manual landmarks is kept to a minimum, this training need not be overly extensive, and the study can be optimized for both reliability and cost.