nicetoolbox.detectors.method_detectors.filters.SGFilter

class nicetoolbox.detectors.method_detectors.filters.SGFilter(window_length, polyorder)[source]

Bases: object

A class designed to apply a 1D Savitzky-Golay filter to smooth and/or differentiate data.

The Savitzky-Golay filter is a digital filter that can smooth or differentiate a set of digital data points by fitting successive subsets of adjacent data points with a low-degree polynomial by the method of linear least squares. This class encapsulates the functionality of the Savitzky-Golay filter, allowing for easy application to data arrays. It is particularly useful for smoothing noisy data while preserving features of the signal such as relative maxima, minima, and width, which are usually flattened by other types of filters.

Implementation note: filtering is vectorized along the frame dimension using scipy.signal.savgol_filter(…, axis=2) on each coordinate slab, rather than nested Python loops over person, camera, and keypoint.

window_length

The length of the filter window (i.e., the number of coefficients). window_length should be a positive odd integer for the intended design; SciPy may adjust the effective length when a series is shorter than the requested window. The size of the window affects the smoothness of the output signal, with larger windows providing smoother results but less sensitivity to small variations in the input data.

Type:

int

polyorder

The order of the polynomial used to fit the samples. polyorder must be less than the effective window length. A higher polynomial order can fit the data more closely, but if too high, it may lead to overfitting, causing artifacts in the filtered signal.

Type:

int

References

Initializes the Savitzky-Golay filter with the specified window length and polynomial order.

Methods

apply

Applies the Savitzky-Golay filter to the input data.
apply(data, is_3d=False)[source]

Applies the Savitzky-Golay filter to the input data.

The filter is applied to each dimension (X, Y, (Z)) of each keypoint separately along time (the frame axis, index 2). The filtered data is stored in a new array.

Parameters:

  • data (np.array) – The input data to be filtered. The shape of the array should be [#Persons, #Cameras, #Frames, #Keypoints, XYZ] for 5D data, or [#Persons, #Cameras, #Frames, XY] for 4D data.

  • is_3d (bool, optional) – A flag indicating whether the input data is 3D or not. If True, the filter will be applied to the Z dimension. If False, the filter will only be applied to the X and Y dimensions. Default is False.

Returns:

The filtered data, same shape as data input parameter.

Return type:

np.array