Isolating sensor orientation w.r.t. gravity and car direction

I have a quandary. I am trying to use MATLAB's IMUfilter code to get orientation. You can get rotation matrices or quaternions. The trouble is: the doc DOES not explain orientation well. Orientation is relative: it has to be respect to something else. I found out the reference MATLAB's imufilter uses: a reference frame (RF) which can either be 'NED' (North-East-Down) or 'ENU' (East-North-Up). the default value is 'NED'. I am trying to get the orientation of a sensor x,y,z, readings w.r.t. to gravity and car direction and I have no idea how to use this orientation result. Do I multiply the accelerometer x,y, and z data by the corresponding orientation matrix? or do I multiply it by its inverse (or transpose). There is a lot of documentation on the MATLAB sensor fusion but I cannot find any useful example along these lines.

[James Tursa]

Do you have some sample data to work with? Can you post some of it? I don't have the Toolbox for this, but maybe we can figure it out from the data.

[Tony Scott]

THX for your interest. Sorry but this data is proprietary. I cannot show it but my question is academic. I have accelerometer and gyroscope readings. Now I have come into contact with Roberto G. Valenti who has been most kind and helpful. MATLAB's sensor fusion IMUfilter can provide orientation w.r.t. ENU or NED. I tried ENU and if I apply the transpose of the 3D orientation matrix (i.e. the inverse rotation) to my accelerometer data vectors, the rotated results point indeed maximally in the negative z (down) direction. That's all well and fine but it seems MATLAB's IMUfilter in itself cannot directly provide the orientation of the sensor w.r.t. the car, i.e. its direction. What is also impractical is that each accel and gyro sample has its orientation quaternion or 3 x 3 rotation matrix... In the end, I want to apply only one rotation matrix to the accel data...

[James Tursa]

If the sensors are constantly changing orientation, how can you expect to use only one rotation matrix for all of it?