New Sensor For Indoor Positional Tracking

[Annon]

Saw this linked from Engadget this morning: http://www.cambridgeconsultants.com/new ... ase/118/en

They claim to use "accelerometers, gyroscopes, magnetometers and pressure sensors" and a bayesian algorithm to track your position with an accuracy of 1% of the distance traveled.

Wonder how responsive it is.

[cybereality]

I saw that too, but they don't really indicate if its tracking in the full 3 dimensions. Could be x/y only, which wouldn't be enough for VR.

[Annon]

I saw that too, but they don't really indicate if its tracking in the full 3 dimensions. Could be x/y only, which wouldn't be enough for VR.

I would imagine the pressure sensor is for Z, but is likely only sensitive enough to pick up you moving from floor to floor.

[geekmaster]

I posted a link to a different study, where they were accurate to about 0.3%, based on recalibrating the zero reference every time the sensor was know to be stationary based on ambulatory gait analysis. They place an IMU on the foot, which is known to be on the ground periodically. In an example test, pickup up and moving an IMU can result in up to 30% position error from double integration of acceleration. But by zeroing velocity at beginning and end of movement, the calculated movement when moving the IMU 1 meter was 0.3 cm (0.3%). The original study used pressure sensors on the foot, but after analyzing all their data, they simplified it to just use an normal IMU, relocated to a foot mount, with NO pressure sensors. They seem to have taken this idea farther than the above lined study in the first post above. I posted a link awhile ago to the more advanced study in a post here somewhere. Need to find the right search keywords to locate it again though...

The reason positional drift is a problem is because a small velocity drift causes large positional error. Zeroing velocity when stopped fixed that, very effectively. And they showed how to detect when stopped for a number of different gaits. As I mentioned in my post, I believe we can to the same velocity recalibration while seated, because we tend to move our heads between viewing positions and then stop our head motion, while we look in the new direction. Zeroing velocity at these times lets us much more accurately calculate postion.

[WiredEarp]

@ Geekmaster: what makes you think they are using that method? Certainly the source link doesn't mention anything about recalibration when stops are detected, nor does it mention a foot based sensor, which would probably be required to guess when the sensor is stationary...

[geekmaster]

@ Geekmaster: what makes you think they are using that method? Certainly the source link doesn't mention anything about recalibration when stops are detected, nor does it mention a foot based sensor, which would probably be required to guess when the sensor is stationary...

Who is this "they" of whom you speak? What source link? Do you mean the link provided in the first post of this thread?

As stated in the FIRST sentence of my post, I was talking about a DIFFERENT study, which was similar to but more comprehensive than this one linked in the first post. The DIFFERENT study I referenced above concluded with more interesting results that may be applicable to the Rift (as I extrapolated for a seated position). We just need to deduce when the Rift HMD is stationary (a pause between gaze redirections), and zero out the velocity component from the IMU data (and reset the fixed "seated upright" position when deduced), resulting in much more accurate postional data from the Rift IMU.

And technically, the pressure pad mentioned in the link from the first post IS a "foot based sensor", and pressure DOES indicate that IT is stationary (but not necessarily the head). A pressure sensor cannot measure distance traveled though, but a foot-mounted IMU can (with 0.3% accuracy according to the DIFFERENT study I refereced).

[WiredEarp]

@ Geekmaster:

As stated in the FIRST sentence of my post, I was talking about a DIFFERENT study, which was similar to but more comprehensive than this one linked in the first post.

Yes, I know. You didn't answer the question. If its a different study, how do you know this one is similar to it? Do you have some further information about the study in the original link? If not, why hijack this post to try to talk about the study you do know about?

the original study used pressure sensors on the foot, but after analyzing all their data, they simplified it to just use an normal IMU, relocated to a foot mount, with NO pressure sensors. ]They seem to have taken this idea farther than the above lined study in the first post above.[/i

Again, what makes you say that, since the link this thread is about doesn't even mention any of this, so why do you think this particular tracker is using the same idea? Unless they are the same people, and just have different versions/iterations of the tech?

And technically, the pressure pad mentioned in the link from the first post IS a "foot based sensor", and pressure DOES indicate that IT is stationary

No-where in that link does it mention feet, and it even says he device concept is small enough to clip on a belt. That doesn't say it IS belt mounted, but it certainly doesn't say its foot mounted. Also, pressure sensor can easily refer to barometric pressure, not a physical pressure sensor.

Finally, as a bit of advice, you should avoid the overuse of caps, when used too frequently, they look unprofessional and make you look a bit excitable.

[MSat]

While 1% error per distance travelled seems quite good for human dead reckoning, it would still add up pretty quick. I also doubt it would be able to detect subtle translations needed for VR.

As for how it works, it seems to operate on a similar principal to Brantlew's Red Rovr, from what I gathered of his description.

[Krenzo]

Yes, I know. You didn't answer the question. If its a different study, how do you know this one is similar to it? Do you have some further information about the study in the original link? If not, why hijack this post to try to talk about the study you do know about?

He mentioned it because it's similar in that it uses an IMU to track position and a filter to minimize drift when the person is stationary.

Here's a picture of the circuit board with labels for what sensors it's using. It just sounds like a refinement of techniques people have been studying like what geekmaster brought up. This might be the study he was talking about.

[WiredEarp]

Thanks Krenzo.

Am I missing something still however? Because I still don't see anything about shoe mounting in that link, or mentioned on the circuit board.

Currently their focus is on standalone units that can be clipped to users

Yep, I definitely dont see any implication that this is shoe mounted. So - where are you guys getting the idea that this is identical to the shoe re-zeroing thing? Seeing as they don't even seem to be the same companies, I'd like to know how you are making the jump from the rather sparse description in the link to talking as though its a certainty that that system uses a foot mounted sensor.
I assume that someone has some more info about the Cambridge system that they aren't mentioning, otherwise it seems like a leap in logic that is a bit unsupported by evidence so far.

[geekmaster]

Yes, I know. You didn't answer the question. If its a different study, how do you know this one is similar to it? Do you have some further information about the study in the original link? If not, why hijack this post to try to talk about the study you do know about?

He mentioned it because it's similar in that it uses an IMU to track position and a filter to minimize drift when the person is stationary.

Here's a picture of the circuit board with labels for what sensors it's using. It just sounds like a refinement of techniques people have been studying like what geekmaster brought up. This might be the study he was talking about.

No, that is a different document than the link I posted in some other thread here (which I cannot find now). My link was to a PDF file too. Thanks for your link, which also looks interesting.

@WiredEarp: The reason for bringing up other RELATED content is so that we can compare and cross-polinate ideas. Yes, I KNOW that I capitalize some words to emphasize them. Is that a BAD thing?

[MSat]

I'm pretty sure this is pretty much an IMU with a barometer (said pressure sensor) for measuring altitude as Annon mentioned (though I have to wonder how well this would work for their suggested use as a firefighter tracker given that heat affects pressure).

[geekmaster]

This looks like the document I was talking about:
http://www.cl.cam.ac.uk/research/dtg/ww ... thesis.pdf

But I found this one linked in a post from brantlew.

[WiredEarp]

@ Geekmaster:

@WiredEarp: The reason for bringing up other RELATED content is so that we can compare and cross-polinate ideas. Yes, I KNOW that I capitalize some words to emphasize them. Is that a BAD thing?

Discussing related content is like capitalization - good when its RELEVANT. In this instance, I have yet to see a link between these two technologies. In fact, we seem to have been sidetracked from discussing what the thread is about (this new Cambridge sensor), to discussing foot based IMU resetting. While this is a good topic, lets not claim we know these two techs are related, if we have no evidence at all to say so.

[geekmaster]

The link in the first post is very sparse (not much info). I interpreted "pressure sensors" to mean exactly what is used to sense pressure between foot and ground. If they were instead meaning barometric pressure (which did not make sense due to thermal issues as mentioned previously), they should have SAID "barometric" or "digital barometer" or something. I thought it WAS related to what I posted. But with such little information provided in that press release, perhaps not...

[WiredEarp]

Mmm, good point. Its possible they were referring to barometric but didn't want to use the term as it might turn non technical people off.

How reliable and quick are barometric pressure sensors anyway? Does the pressure change when moving? If so, I wonder if it would be possible to somehow detect being stationary using a sensor like that.

[MSat]

Mmm, good point. Its possible they were referring to barometric but didn't want to use the term as it might turn non technical people off.

How reliable and quick are barometric pressure sensors anyway? Does the pressure change when moving? If so, I wonder if it would be possible to somehow detect being stationary using a sensor like that.

Barometric sensors are more than quick enough to determine altitude changes. Some Android phones and tablets now use them to help get a faster GPS lock (as now they could at least estimate their altitude).

A barometer could probably determine if you're moving in an elevator, or walking up stairs (in conjunction with the other sensors), but it wouldn't be of much use when walking on a flat plane - well, other than determining that you're indeed walking on a flat plane.

A barometer is vital to determine what floor someone is on - particularly when the area it's being used in has not been mapped.

Regardless, this method is likely of limited use for VR like I mentioned earlier.

[android78]

I don't quite see the point of the pressure sensor. They are reasonable accurate, but the problem is that the pressure changes due to weather can cause a drift of several meters within a short amount of time. I guess it can (generally) tell if you're moving up or down, but I don't think this would be any more accurate then just having an accelerometer.

[MSat]

Well, this has to have some sort of recalibrate function, as it'll be off by a foot every 100 feet. During that time, I assume the altitude would also be recalibrated.

[nanicoar]

There is something called Differential GPS which is known to achieve down to 10cm accuracy outdoors. Indoors there is a problem called 'multipath' where the receiver is unable to distinguish signals from the reflection of said signals, but this can largely be compensated for in software. - Even a rather dinky accelerometer could be usable for position tracking if it's corrected with accuracy like that of DGPS.

If what I have understood what I have been reading correctly, DGPS works by sending correction information from a fixed position that you know the location of down to pinpoint accuracy. To this correction information you can add measurements of atmospheric and ionospheric disturbance and that is when accuracy goes below one meter. All of this is transmitted at a low frequency of around 300kHz, and if you live near a port you are very likely to constantly have this high-quality data in the air... sent at just around 100 bits per second. You could pick it up with the antenna from an AM radio, or just coil up the antenna yourself.

AFAIK there is no reason why this DGPS correction information can't be sent over the internet, no reason why it can't be used on an A-GPS smartphone, or a USB GPS dongle, or something that you can stick in the back of the headband of a HMD like the Rift. There already exists software such as http://www.gpstk.org/ which can knit this data together.

As such, I have strayed just a little from developing my smoothing algorithm for the magnetometer in my phone... >_>
But I found the API documentation for accessing raw GPS measurement data in my smartphone and I have already made a programs that transmit raw sensor data to my PC. A PC has the horsepower to knit this all together in realtime. It's doable. I have to try!

[Krenzo]

Indoors there is a problem called 'multipath' where the receiver is unable to distinguish signals from the reflection of said signals, but this can largely be compensated for in software.

Oh really? Would you mind pointing out some research papers that explain how to compensate for it in software? I would think that's not possible since that causes a loss of signal in the hardware side before the signal even reaches the software.

[nanicoar]

Oh really? Would you mind pointing out some research papers that explain how to compensate for it in software? I would think that's not possible since that causes a loss of signal in the hardware side before the signal even reaches the software.

"With House holder Sliding Window RLS ¯lter, error is reduced by 68.5% and 81.9% with convergence time of 0.69 hrs and 0.39 hrs for MP1 and MP2 (Figures 11 and 12) respectively."
-- http://www.jpier.org/PIERM/pierm21/10.11080811.pdf

But actually I said this because I saw a mention of multipath in the Symbian C++ documentation. The above is just some quick googling.