Kalman filters, gyros and accelerometers in 3DOF, Oh my!

1. Oct 5, 2008

Ian_Brooks

Can someone please explain to me how kalman filters would help a system with accelerometers and Gyros work better together?

Thanks

2. Oct 6, 2008

Staff: Mentor

Can you be more specific? What is the source of the assertion that Kalman filters would provide special help?

3. Oct 8, 2008

Ian_Brooks

sure -

I know a kalman filter uses a predictive feedback method by comparing a known reference predicted value to an actual value measured in noisy environments. If I use a system that uses gyroscopes and accellerometers for Innertial measurement, I was told a Kalman filter bridges the gap between sensory data and the actual position.

However, without knowing exactly how a kalman filter would solve my problem i'm having a little trouble getting started.

i'm currently reading the following paper to see where I need to go next.

http://www.invensense.com/shared/pdf/motion_sensing_accel_human.pdf [Broken]

Last edited by a moderator: May 3, 2017
4. Oct 9, 2008

Topher925

Kalman filters are a good way of combining the best of both worlds of the gyroscope and accelerometers. Gyroscopes work by integrating constantly over a period of time, which creates error because it is an iterative process. This is usually referred to as "drift". Accelerometers measure acceleration in only one axis and typically don't generate and error. However they do measure all acceleration, this includes the gravitational field and what ever acceleration the sensor may be experiencing. A kalman filter is often used to tie the two together, the gyro measuring rotation, and the accel minimizing the error that the gyro creates (accel is offset from center of rotation, purhaps delta pattern). This doesn't have to be done with a kalman filter, there are a lot of easier ways of doing it but they don't perform as well.

There are a lot of different kinds of kalman filters. The most common is the discrete kalman filter which is used for linear systems and normal or gaussian distributions. There is also the extended kalman filter typically used for stochastic non-linear systems. Then there is also different breads and hybrids like the unscented kalman or kalman-bucy filter.

I suggest you start with the discrete kalman filter as it is the simplest and easiest to implement. If your writing code for this project, a discrete kalman filter can usually be implemented in only 5 lines of code after the system is defined.

This is a great paper if your just getting started with the KF: http://www.cs.unc.edu/~welch/media/pdf/kalman_intro.pdf

I needed to implement an extended kalman filter for a non-linear system for a project at work and this got me up to speed very fast.

5. Oct 12, 2008

Ian_Brooks

thanks, i'm currently making a hardware protoype with the sensors mounted onto a single board, the software implementation will come into play next semester.

6. Oct 2, 2011

sam15

Hi,

If I only have accelreometer readings and no gyroscope or orientation readings is it still possible to filter out g from it using a kalman filter? Also will kalman filter help me reduce the noise from the accelerometer data better than a moving average or a low pass filter?