Can 9DOF IMU tech ~accurately locate an athlete in an enclosed space

Hi,
Realising this is a return to a topic quite well hashed out since 2011 or so, I apologise in advance.

However, now, 2017, after the advent of the ubiquitous Quadcopter revolution I'm wondering if any recent advances in the tech have made it possible to track and locate a moving person carring/wearing a 9DOF IMU with a reasonable degree of accuracy, from a standing start to completion of stop-start movement over a period of, say, 15 minutes.

The idea is that the sensor pack would be given to the wearer just prior to commencement of the 15min, and the data collected wirelessly during his/her progress through a course. Data points - spatial locations - in a plane are what I'm looking for (i.e. a grid reference) whereby actions taken at those locations have a direct bearing on the "progress" being made by the athlete.

The person will be walking or running from point to point and then standing - always standing but not likely immobile - when the waypoint is measured, triggered by the athlete hitting a button on the pack, which in turn triggers the action at that waypoint. The athlete chooses the point he/she stops at to carry out the task, so the distance from the "goal" they are going for is the variable I need to measure.
Example:
Alan starts at START and runs toward "Goal1": When he gets to where he feels he can accomplish his task (think throwing a ball through a hoop for an example) he stops, hits the button and the "goal" becomes available to him. If he's 2 meters away he scores less than if he's 5m away. Hence the need for the positional data at the button press. "Goal1" completed he then proceeds ,walking or running (his choice as speed v accuracy becomes a factor) to "Goal2". Rinse, repeat.

Accuracy of 1 metre over a 50x50m course would do - can I get near to this, or better?

Has anyone managed to nail this kind of analysis with any degree of success?

Due to the dynamic rates of change by an active person, i suspect a low-end Arduino might not be able to keep up in real time to maintain 3D spatial awareness/integrity over 15 mins straight.
Interesting project.

from a standing start to completion of stop-start movement over a period of, say, 15 minutes.

Not a ghost of a chance. Affordable technology hasn't advanced that much.

Read more about why not, here: http://www.chrobotics.com/library/accel-position-velocity

Pozyx, which is based on trilateration, might be usable, but it is not cheap.

Thank you for your replies.
@Lastchancename - yes, it is proving very ...interesting indeed.
I had thought of using RFID tagging and hard stations to each task, but where's the real fun in that? The concept of permitting the competitor use their own judgement and pick their own comfort zone really appeals. It is the implementation of this that is causing me the real problems.

@Jremington - I really hope you are wrong in the assessment "Not a ghost of a chance." I had read the chrobotics article carefully (why, oh why, will people not date their articles???) and concluded it was not recent, and that the advances made of late in drone racing might just possibly have yielded some new approaches.

cm accuracy is not needed for my project, and bearing in mind that the time interval is relatively low overall my question still stands: Is there anyone who has had any success in implementing a cost-effective tracking system that I could benefit from emulating?

Could I hybridize? Is there value in considering placing a localizer at each "goal" and re-setting the tracker throughout the course as each goal is encountered?

Ideas?

You ask for 9DOF... is that necessary for all 9 axes, or are you looking for any subsets of XYZ position, heading, rotation?
When you eliminate any one of these, it reduces the maths complexity by an order of magnitude.

@lastchancename
In reality just the xy coordinated of the player, and the time, at the chosen trigger point - or, more accurately, at the point played from after the trigger moment in case of trigger-move-play.
I asked re the 9dof as I was looking for a way to get some accuracy into an otherwise notoriously inaccurate system if reports are to be credited.
Mapping the players' movements would have been a cool side-effect which I had planned to utilise later in the dev by permitting (or forcing) the player to choose the route between as well as the order of the goals.

This project is focused on youth development, and I hope it will be possible to give good gaming through the application of these ideas - combining the mental - through decision making and the physical with confidence building.

A completely different proposition than the title.
You've eliminated 7 of the 9 axes for the first iteration - a much better starting point. Your calculations are now literally 10 million times less intensive!
How often do you need to update the position? You may well have enough time to do XY tracking in near real-time.
To maintain accuracy over a longer period, you may have RF or IR beacons in the play field, to recalibrate any single access whenever the player crosses that 'virtual' line.
When you get rev.1 sorted, you can think about more power, axes and accuracy.
A few years ago, my colleague tried something similar with trilateration of RFID tags indoors, but with the 'bases' about 20' apart, we could only get around 2' accuracy, with wildly varying margins of error (irregular reflection and absorption of the RF within the area of interest). Using these newer chips is a great idea, and something like BLE to send the data back, might give you a chance to write your comped on a larger/desktop style machine.

Full on aviation grade IMUs drift on the order of a kilometer in an hour. A meter in 15 minutes isn't going to happen.* Drone IMUs use a combined IMU + GPS solution where the GPS effectively limits long term drift and the IMU captures instantaneous dynamics. With that sort of system 1 meter typical error might be plausible, but it won't work at all indoors.

There are commercial systems that use video tracking to record player movements and I've seen reports of projects doing this sort of thing on platforms like Raspberry Pi. If it were my project, that's what I would look at first.

    • Back of the envelop analysis:
  1. 15 minutes is 900 seconds
  2. Position is double integral of time (acceleration -> velocity -> position) so acceptable error limit is 1 meter / (900 seconds)^2 or about 1e-6 m/s^2.
  3. ADXL345 for example can resolve at best 13 bits so +/- 2G/2^13 or about 2e-4 m/s^2 and that's probably quite optimistic.
  4. Ignoring all other potential errors, accelerometer is at least two orders of magnitude worse than your requirement.

@Jremington - I really hope you are wrong in the assessment "Not a ghost of a chance." I had read the chrobotics article carefully (why, oh why, will people not date their articles???) and concluded it was not recent, and that the advances made of late in drone racing might just possibly have yielded some new approaches.

Your question comes up all the time, so lots of people in this forum and elsewhere would be absolutely delighted if you could get the project to work to your satisfaction.

Go for it, and please keep us informed!

Thank you all for your input - and especially for the ruined envelope, very instructive @MrMark :slight_smile:

I most certainly will keep you all informed of my progress ( or lack thereof ) as I stagger forward.

The BLE concept is of merit and worth researching further @lastchancename; have you any particular kit you'd recommend?
Your colleague's trilateration concept: RFID is really short range, no? - how did he propose to get workable distances?

@jremington: Posyx - w$w! Seems like a solution, but superleague.

I am well familiar with the whole Video Analytics field, and had rejected it early on for portability reasons. There is a need for this system to be portable and easy set-up. Good idea though @MrMark.

The RFID solution was scaled down to 20x20m for demo - but the error was still outside ypur requirements.
These were an active tags (Australian brand - I'd have to look), but our design was not for high precision - so the 2m error was OK.
Thge actual prototype was developed for approx 500x500m - so well within your spec,

lastchancename:
Thge actual prototype was developed for approx 500x500m - so well within your spec,

What did you use for the actual prototype, and what error did you stay within? I'd be interested in knowing more...

I just heard back from the fellow t hat did the code

""it was Protrac ID from Queensland.
But if I was doing it again, I would use Bluetooth 4 or Wirepass.
We recently tracked BT beacons around our office using this method and it will prove a lot less expensive to deploy. ""

Good luck.