dispel.signal.accelerometer module#
Accelerometer functionality for signal processing tasks.
- dispel.signal.accelerometer.GRAVITY_CONSTANT = 9.80665#
The gravitational acceleration near Earth’s surface.
- dispel.signal.accelerometer.apply_rotation_matrices(rotation_matrices, sensor)[source]#
Apply rotation matrices on a sensor time series.
- Parameters:
rotation_matrices (Series) – The rotation matrices obtained with
compute_rotation_matrices()sensor (DataFrame) – The sensor time series to be rotated
- Returns:
The rotated sensor values based on
rotation_matrices.- Return type:
- dispel.signal.accelerometer.compute_quaternion_between_vectors(v_1, v_2)[source]#
Compute quaternion given two vectors.
- Parameters:
- Returns:
numpy.ndarray – An array containing the quaternion of shape (n_samples, 4).
See proposed solution with pseudocode here
https (//stackoverflow.com/questions/1171849/finding-quaternion-)
representing-the-rotation-from-one-vector-to-another
- Return type:
- dispel.signal.accelerometer.compute_rotation_matrices(gravity, target_gravity)[source]#
Compute rotation matrices based on gravity time series.
- dispel.signal.accelerometer.compute_rotation_matrices_quaternion(gravity, target_gravity)[source]#
Compute rotation matrices from gravity time series (quaternion-based).
- dispel.signal.accelerometer.dot_diag_einsum(v_1, v_2)[source]#
Get the diagonal of the dot product of two 2D arrays (fast).
This function is based on np.einsum configured in such a way to provide the dot product diagonal of two 2D arrays. For benchmark example (n_samples=18000), runtime is 2 ms.
- Parameters:
- Returns:
An array of shape (n_samples, 1) containing the dot product diagonal
- Return type:
- dispel.signal.accelerometer.dot_diag_list(v_1, v_2)[source]#
Get the diagonal of the dot product of two 2D arrays (slow).
This approach computes the dot product for each time sample of the arrays of shape (n_samples, 3). For benchmark example (n_samples=18000), runtime is 60 ms.
- Parameters:
- Returns:
An array of shape (n_samples, 1) containing the dot product diagonal
- Return type:
- dispel.signal.accelerometer.dot_diag_matrix(v_1, v_2)[source]#
Get the diagonal of the dot product of two 2D arrays (very slow).
This approach computes the dot product array of two arrays of shape (n_samples, 3) and then takes the diagonal. For benchmark example (n_samples=18000), runtime is 2 seconds.
- Parameters:
- Returns:
An array of shape (n_samples, 1) containing the dot product diagonal
- Return type:
- dispel.signal.accelerometer.orthogonal(v_1)[source]#
Find orthogonal quaternion to a vector provided.
Based on C++ implementation here: PX4/PX4-Matrix (lines 192-224)
- Parameters:
v_1 – An np.ndarray of shape (3, )
- Returns:
An array of shape (4,) containing the 180 deg quaternion rotation
- Return type:
- dispel.signal.accelerometer.quaternion_arr_normalize(q_arr)[source]#
Normalize an array of quaternions.
- Parameters:
q_arr (ndarray) – A numpy.ndarray of shape (n_samples, 4) containing the quaternions.
- Returns:
The unit quaternion in the same format as the input.
- Return type:
- dispel.signal.accelerometer.quaternion_rotate_vector(q_ba_np, v_a_np)[source]#
Rotate a time series of a 3d vector by a quaternion.
- Parameters:
- Returns:
A numpy.ndarray of shape (n_samples, 3) containing the vector, expressed in coordinate frame b.
- Return type:
- dispel.signal.accelerometer.remove_gravity_component(data)[source]#
Remove the gravity component of acceleration.
Based on paper : Two-stage Recognition of Raw Acceleration Signals for 3-D Gesture-Understanding Cell Phones, Cho et al., 2006
Get the linear accelerations - without gravity component - by subtracting the mean acceleration signals, such that \(A_1(t) = A(t) - A_mean where A(t) = [a_x(t),a_y(t),a_z(t)]\)
- Parameters:
data (DataFrame) – Input acceleration data
- Returns:
A tuple with first, the acceleration data with mean removed and second the mean.
- Return type:
Tuple[pandas.DataFrame, float]
- dispel.signal.accelerometer.remove_gravity_component_ori(acc_sensor, q_global_sensor, unit='g')[source]#
Remove the gravity component of acceleration based on orientation.
Get the linear accelerations - without gravity component - by converting to the global coordinate frame, subtracting the constant gravity and converting back to the initial sensor frame.
- Parameters:
acc_sensor – Input acceleration data expressed on the sensor frame.
q_global_sensor – Input orientation from sensor to global coordinate frame.
unit (str) – The unit in which the accelerometer data is expressed.
- Returns:
A tuple with first, the acceleration data with mean removed and second the gravity.
- Return type:
Tuple[pandas.DataFrame, pandas.DataFrame]