I'm wondering if we can measure 3-D length only by our smart-phone's accelerometer. And we all know these low cost IMUs are not accurate.

You can model accelerometer's error this way: $$ a = f*a' + g + b + \eta $$

where $a$ is the actual acceleration, $f$ is a 3x3 matrix modeling scaling, misalignments, cross-axis and ... errors. $g$ is gravity, $b$ is bias and $\eta$ is noise.

I'v found a good method to calibrate(compute $f$ and $b$) smart-phone's accelerometer: Estimate smartphone accelerometer bias

Now I have to:

1) remove $g$

2) remove $\eta$

How can I do that accurately? As I mentioned above, this problem requires good accuracy.

If you believe accelerometers are not accurate enough for this problem, and other sensors should be used (Sensor Fusion), please share your idea.

  • $\begingroup$ How do you intend to use the accelerometer for 3D lengths? $\endgroup$ – Cherny Jun 17 '18 at 10:42
  • $\begingroup$ By moving smartphone and compute displacement from acceleration information. @Cherny $\endgroup$ – HsnVahedi Jun 17 '18 at 10:49
  • $\begingroup$ So I don't believe it's accurate enough since your measurement is an integral of the acceleration, and it usually turns out highly inaccurate. Usually sensor fusion is used for this purpose, but the accelerometer is only relevant because it can measure very frequently $\endgroup$ – Cherny Jun 17 '18 at 10:50
  • $\begingroup$ Thanks! I agree with you. It seems accelerometer is not enough for computing 3-D displacement. But what about 1-D displacement? I mean you move accelerometer in a line. By knowing that the sensor's data instances are collected when the movement was 1-D line, Can you correct the data and estimate the 1-D displacement?@Cherny $\endgroup$ – HsnVahedi Jun 17 '18 at 11:02
  • $\begingroup$ Welcome! you mean project the estimated 3D location on the 1D line? if so then I believe it's still about the same error, since projection will just give you the 1D error $\endgroup$ – Cherny Jun 17 '18 at 11:05

Your model may be misleading, as bias and noise (including non-linearity and quantization noise) get double integrated, thus error increases quadratically and without bound over time.

  • $\begingroup$ Thanks for your answer! This article: ieeexplore.ieee.org/document/5594974 , suggests a method for dynamic bias and gravity estimation. So the time-dependent bias parameter can be estimated dynamically. And I hope there exists some way to reduce noise effect. $\endgroup$ – HsnVahedi Jun 17 '18 at 18:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.