#052 IMU motion tracking: Why is it so hard? // Tutorial


Hi. This is a continuation of my earlier
videos on motion tracking technology. In this video we’re going to be taking a
look at why IMU tracking is really hard to do in certain situations.
For example the gym. I’m a real gym junkie.
If I haven’t been to the gym
at least five times in a week then there’s something really wrong,
but i also love technology. What would be really nice; if some
technology, some usable technology, could be put into the gym this is one of my goals but i found some
technology just really isn’t up to it. So to show you what I mean I took an IMU suit
into a real-world scenario. A good buddy of mine has a gym called
“Clean Shred” out at Castle Hill so I’ve come along because I want to see how I can get
some motion capture units going for the fitness industry.
So, let’s go in and see how he’s going. . Here he is!
He’s over here. Simon! How are you going, mate? You again!
I thought I told you to never come back down here!
Get outta here! Er. OK.
We’ll come back to Simon later. So, several years ago I started coding
up an IMU tracker with the express purpose of being able to use it in the gym.
It had to keep track of:
How many sets and reps i did.
How quickly i did them. How much rest I had between sets.
How much effort I was expanding. and How my form was going.
So the hitch-hiker was born.
It was just a Simblee with a 9150 IMU attached. I wrapped it up
with electrical tape and put a magnet inside so I could easily attach it to a
barbel but I was getting frustrated with the data coming out of it.
I had to apply lots of algorithms such as data smoothing, peak detection, predictive analysis
and other fuzzy bits to ignore such things as dropping dumbbells on the ground.
So I got it to the stage where it was pretty accurate but it still wasn’t good enough.
So I took a look at what had progressed in the industry since i last looked.
It seems that there’s still not any proven technology that helps the
weightlifter but there’s some new stuff for example: There’s Push which is a motion
tracker attached to your wrist same old, same old.
and Atlas have a range of wearables with, once again, IMUs in-built but now we’re starting to see things
like cable based trackers. For example Kinetic and also OpenBarbell!
Cool! I can imagine cables will be fairly accurate So why is IMU technology really
useless for weight lifting? There’s fundamental reasons. Noise and drift.
For this experiment I used the Kickstarter I backed last year called Perception Neuron.
This is a full motion tracking suit that is very accurate but offer requires
someone to fix up the data when using it for motion capture in videos. Let’s take a fairly basic tracking task
such as running. That actually looks quite good, doesn’t it? The red balls appearing at my feet show the
current zero point location for height, and the big red ball shows my center of
gravity. Now what about doing some pull-ups?
The software will assume that my feet are always touching the ground and will
result in some rather bizarre activities. You can fix this up after capturing but
a little pointless for my needs. Deadlifts are fairly easily to capture because my
feet are always on the floor. You can see that the capture is pretty good, even
though my form is hopeless.
Bad form! Naughty form!
Taken from another angle
you can see certain parts of my anatomy drifting. For example; my feet were
straight doing these deadlifts but my left foot on the capture was pointing in. Doing let pull-downs was OK. Actually it
was a pretty good capture and the rowing machine was coming up with a decent
capture as well, but leg presses… Well This is where it starts to get a little funky.
The software still thinks my feet are firmly on the ground and all I’m doing
is pushing myself into the ground. Can you figure out what i’m doing here?
No, not a backflip push-up thingy! It’s supposed to be a barbell bench press.
This is a great example of where the first problem lies and it’s called drift.
IMU drift is caused by the simple fact that each sensor does not have a
real-world frame of reference. All they’re doing is guessing where they are
in a real world based on how much acceleration they experience.
This is called “dead reckoning”.
This guesswork becomes even worse when you take into
account errors produced by noise, temperature and scaling issues.
Common ways of dealing with drift is to apply a frame onto the sensors called an amateur. This limits their drift away from other sensors.
This is what the Perception Neuron software does but as you saw
in my examples it still gets it slightly wrong. Another way of dealing with drift is to
have a real-world beacon in the room where the army suit can triangulate its
my position and then all the other sensors can either be brought into line
using an amateur, or have every sensor trianglulate, or have no sensors and
just use cameras instead. This is expensive, still prone to errors
and not particularly portable. Although, new technologies coming from
DARPA look promising. The second problem with IMUs is noise.
Some of the exercises I did came up with some great captures like the dumbbell
bench fly.
Notice I used a different bench? This bench had less metal in it
than the previous bench. Which had a lot of metal in it. You can see what happens
when I do a barbell squat.
A big lump of metal placed on my shoulders, with the
big metal squat rack nearby, makes my squats look like ballet routine.
Very dainty.
So why is this happening? Well it’s all about magnets. The earth is
one big magnet with a north and a South Pole. The IMU suit will make use of this
as a crude triangulation method. It’s actually fairly accurate but the
magnetic field is so faint that large metal objects will affect the sensor readings.
Not only metal objects but phones, computers, mains power and anything
else that emits electromagnetic waves. In fact the Perception Neuron guys are so
worried about it that they provide a magnetically shielded case to house the sensors.
Going back to my original goal. How can I automate my gym session
tracking and gather some performance data? As I showed you earlier there’s
some really interesting progress made with cables but that’s a little bit bulky.
But there’s two other ways I’m thinking of to achieve my goals. Stay tuned next time to find out what
they are and how my test results went. Oh by the way. If you live around the Castle Hill
area, check out Simon’s gym. Actually it’s not just a gym. They train anyone
for beginners to pro athletes and he’s always looking for new ways to train.
So, if you’re in the area check it out! So what do you reckon about this?
Hmm? It’s a pretty good gym, isn’t it?
If you enjoyed this video then don’t forget to like and if you’re not a subscriber, it’d
be great to see you as one.
So see you next week. 99… 100…

Leave a Reply

Your email address will not be published. Required fields are marked *