Android：有什么方法可以准确区分电话运动和旋转吗？

Question

I'm currently working on a VR-Framework. My Android Smartphone is used as stereoscopic display and controller. I have headtracking working, but I want a simple positional tracking, too. I know that a precise tracking of the smartphones position is not possible. At least as long as one is limited to the smartphone's sensors. So the idea is to distinguish between a rotation of the smartphone (headtracking) and just any type of general motion. Of course with the consequence that I'm just able to move eg forward if motion is detected. As I said, headtracking works fine with the android gyroscope. But the motion tracking is still not working as I want it to be.

The main problem is that when rotating the phone, this rotation is actually also an acceleration. So just looking at the acceleration's magnitude is not good enough to distinguish between acceleration(motion) and rotation. Furthermore, doing something like a threshold for the gyroscope:

if(gyroscope.magnitude < EPSILON)
    if(accelerometer.magnitude > EPSILON)
        //this may be motion

does not yield the desired differentiation either, because there is always a degree of rotation going on with the phone while walking.

Long story short: Is there any possible solution for this problem? (Edit) And if yes, how is it done? I want to be able to distinguish if the smartphone was moved or rotated. Ideally only with the usage of the gyroscope and accelerometer.

Best regards and thanks for any feedback!

Answer 1

I'd suggest to get familiar with the concept of quaternions and how they can be used to represent orientations (This might be a good start: http://run.usc.edu/cs520-s12/quaternions/quaternions-cs520.pdf )
Also using heuristics as the "EPSILON" you suggest might not be sufficient since there are maybe some cases in which the heuristic fails but Im sure you know that already.
Most of the time you have to care about bias in the raw data therefore you either have to make sure that the phone is at rest for a sufficient amount of time (depends on the number of sample you get in that time period = depends on the sample rate of your sensors) at the beginning of your measurement. In that case you could calculate a mean on the axes and substract it. This assumes to the bias to be constant over time and does not care fo bias instability (Take a look here: http://sensorwiki.org/doku.php/sensors/gyroscope ). Bias instability doesnt only apply to gyros.
In case you cant guarantee rest at the beginning of the process you also want to read something about Kalman Filters since they provide you with the possibility of estimating the bias. I've seen implementations of KF using the quaternions and the bias as a state ;) - just for starters.