Get Rotation Matrices

[CarlosGomezHuelamo-ATAI]

Hi!

This dataset is quite interesting. I am trying to explore different algorithms to estimate driver behaviour using inertial data, and I found that this could be a potential dataset. Nevertheless, in order to determine the car's acceleration based on the accelerometer, gyroscope etc. first of all we need to find out different rotation matrices, such as the one from the car reference frame to the sensor reference frame, and the one from the sensor reference frame to the world reference frame, in other words:

R_CS (Car -> Sensor)

and

R_SW (Sensor -> World)

This can be done using maths, but apart from that, did you record it using the Android App?

On the other hand, was the smartphone mounted during the whole trip? So the rotation is constant between the car and the smartphone.

Looking forward to hearing from you.

[jair-jr]

Hi!
Thank you for your interest in the driver behaviour dataset.

did you record it using the Android App?

Yes. Here is the app: AndroidStudioProjects/AnalisePerfilMotoristas_v1. It's a bit outdated, but I think it still works.

was the smartphone mounted during the whole trip? So the rotation is constant between the car and the smartphone.

Yes, the smartphone was mounted on the panel of the car during the whole trip. So, we can assume that whatever happened to the car also happened to the phone.

All the best,
Jair Ferreira

[CarlosGomezHuelamo-ATAI]

Thanks @jair-jr @Brunovf1. I have seen that the sensors you provided:

Nevertheless, these values are not in sensor frame (we expect x-left, y-up, z-forward) but in car frame (x-forward, y-left, z-up, which is like a LiDAR frame), since in Raw Acceleration, the gravity is acting in the z-axis. Can you report how did you make these transformations and if you have the original values to validate? It must be in the code, but I could not find it.

[jair-jr]

Hi Carlos.

Yes, you are right. I transformed from the smartphone coordinate system to the Earth coordinate system. This was done by multiplying a rotation matrix (3 x 3) by the sensor matrix (3 x 1). The rotation matrix is provided by the Android API. Specifically, by a method in the SensorManager class: getRotationMatrix.

Unfortunately, I don't have the original values as I only stored the transformed values on the device. I did not store the rotation matrix either, which would be necessary to perform the transformation from raw sensor data.

Here is the class that performs the conversion: EarthAxesConverterFunction

Here is a test case of this class that runs on the device: EarthAxesConverterFunction2Test

Here is a test case of this class that runs on the computer: EarthAxesConverterFunctionTest

All the best,
Jair Ferreira

[CarlosGomezHuelamo-ATAI]

Hi Jair. And after the conversion, where are the axis pointing?