If You Care About Accuracy, Your 3D Body Scanner Needs to Have Handles

handle-anything.png

We did a short interview with Fit3D Greg Moore on the physical design of Fit3D & what decisions were made to make it as accurate as possible. There were 3 big factors that mattered if you wanted an accurate system:

  1. A quality 3D Depth sensing camera must be used.

  2. Handles are needed to reduce movement by a significant factor. Also serve as a safety measure.

  3. The right speed must be chosen for an accurate scan.

For background, when I, Patrick, started here at Fit3D, I asked Greg how they designed our system to maximize for quality. It turns out we did quite a bunch of research before we came out with our first production unit a several years ago. Here are the biggest learnings.


1. Camera:
Some brands use a Kinect camera that you can literally get off eBay for under $20 dollars. See the search results here (LINK). While this does not exactly match price to quality, in general we can draw some connection to off the shelf hardware that is “hacked” to do something else. If your entire system is based off a camera that is used to play games for a 10 year old, that would give me pause. Plus, it’s no longer produced by Microsoft (LINK), which gives me another pause. Now, moving on to the data experiments.

2. Handles: Greg reported that when we started the design process they started with handles & without handles. We did one test spinning at 30 seconds of a unit with handles & without handles. Our hardware unit without handles experienced consistently twice (2x) as much movement as the hardware unit with handles! That’s a lot of movement to consider when a camera is capturing loads of data every second & trying to construct a 3d model of your body.

If you care about accuracy, you want to minimize movement. Any movement results in noise of the scan, which asks the computer to guess your measurements.

You can give this a try at home: Hold your arms straight out & look at your finger tips (spread apart) for about 10 seconds. There is a good chance that you will see your fingers moving slightly up & down. Maybe even your arm? Imagine taking a photo from 30 inches away at intervals of 1 second & 2 seconds. Then you need to average those 2 points together in 3D space. These might seem like little movements, but they are big movements if you consider they are the foundation for your scans & your mapping the human body.

There is one other important safety consideration about handles. Imagine a 60 year old client who’s balance is poor & perhaps they are overweight. Now imagine asking them to stand still on a rotating platform without falling off as it spins. Does this sounds safe to you? Visualize for a minute that moment when the platform is spinning and think about what happens if they fall off.

3. Speed: The last item we mentioned was speed. We did an additional test where we tried to see if we could scan a person at 17 seconds instead of 30. Why not? Quicker is better, right? What we found was that capturing a person’s 3d body scan had 4x more movement without handles & at 17 seconds. Thus, 17seconds was too fast to consistently deliver an accurate scan.

All these small experiments might seem small, but at Fit3D we to deliver the best 3D body scans possible. This is why we have handles & run all our machines at a 35 second scan time.

While I am sure that you could cut some corners & do without these features, but after you know the science behind how to minimize any error in your system, why would you settle for less?

Patrick Donnelly