skdh.utility.orientation.correct_accelerometer_orientation#
- skdh.utility.orientation.correct_accelerometer_orientation(accel, v_axis=None, ap_axis=None)#
Applies the correction for acceleration from [1] to better align acceleration with the human body anatomical axes. This correction requires that the original device measuring accleration is somewhat closely aligned with the anatomical axes already, due to required assumptions. Quality of the correction will degrade the farther from aligned the input acceleration is.
- Parameters:
- accelnumpy.ndarray
(N, 3) array of acceleration values, in units of “g”.
- v_axis{None, int}, optional
Vertical axis for accel. If not provided (default of None), this will be guessed as the axis with the largest mean value.
- ap_axis{None, int}, optional
Anterior-posterior axis for accel. If not provided (default of None), the ML and AP axes will not be picked. This will have a slight effect on the correction.
- Returns:
- co_accelnumpy.ndarray
(N, 3) array of acceleration with best alignment to the human anatomical axes
Notes
If v_axis is not provided (None), it is guessed as the largest mean valued axis (absolute value). While this should work for most cases, it will fail if there is significant acceleration in the non-vertical axes. As such, if there are very large accelerations present, this value should be provided.
If ap_axis is not provided, it is guessed as the axis with the most similar autocovariance to the vertical axes.
The correction algorithm from [1] starts by using simple trigonometric identities to correct the measured acceleration per
\[a_A = a_a\cos{\theta_a} - sign(a_v)a_v\sin{\theta_a} a_V' = sign(a_v)a_a\sin{\theta_a} + a_v\cos{\theta_a} a_M = a_m\cos{\theta_m} - sign(a_v)a_V'\sin{\theta_m} a_V = sign(a_v)a_m\sin{\theta_m} + a_V'\cos{\theta_m}\]where $a_i$ is the measured $i$ direction acceleration, $a_I$ is the corrected $I$ direction acceleration ($i/I=[a/A, m/M, v/V]$, $a$ is anterior-posterior, $m$ is medial-lateral, and $v$ is vertical), $a_V’$ is a provisional estimate of the corrected vertical acceleration. $theta_{a/m}$ are the angles between the measured AP and ML axes and the horizontal plane.
Through some manipulation, [1] arrives at the simplification that best estimates for these angles per
\[\sin{\theta_a} = \bar{a}_a \sin{\theta_m} = \bar{a}_m\]This is the part of the step that requires acceleration to be in “g”, as well as mostly already aligned. If significantly out of alignment, then this small-angle relationship with sine starts to fall apart, and the correction will not be as appropriate.
References