Does the signal processing block normalize the data before generating 33 features for 3-axis accelerometer data?
@naveen: It does two things:
- Remove the mean from the signal.
- Apply a filter if set.
Typically enough to normalize, but it doesn’t normalize to 0…1
Thanks @janjongboom for letting me know. To be clear, I am planning to use 3-axis accelerometer data with 3-axis gyroscope data. Since they both have different units and ranges I was assuming the DSP block would handle that automatically but I guess I would need to normalize them before uploading to the Edge Impulse. Just thinking if only subtracting mean will make them on the same scale?
@naveen use two different spectral features blocks, one for accelerometer and one for gyro, then use the ‘Scale’ field in each block to scale to the same range (e.g.
0.001 if you want to divide gyro data by 1000).
@janjongboom Thanks for the pointer! I will try it soon.
@janjongboom I think Scale field cannot scale the two spectral blocks say acc vs gyro on the same scale. The mean which is subtracted from the data is local mean and the scale value is global (could be 1/std or 1/(max-min)) so it would not work.