ibo-biomech

Python tools for motion-capture workflows: load C3D and the institute's HDF5 format, process markers and signals, export OpenSim inputs, and inspect IK/ID results. Both file handlers return a TrialData containing named channels.

The package is in alpha. See the current limitations and development priorities for the status of HDF5 saving, force resampling, and coordinate assumptions.

Install

python -m pip install ibo-biomech

For this checkout, including the latest local changes:

python -m pip install -e .

Plotting methods require matplotlib, installed separately:

python -m pip install matplotlib

OpenSim is only required to run scaling, inverse kinematics, or inverse dynamics. File conversion and result containers work without it. Follow the OpenSim Python setup instructions and verify import opensim in the same Python environment.

Try a complete example without a data file

This example creates a marker sampled at 100 Hz, filters it, crops a copy, and exports a DataFrame with timestamps. The cutoff is illustrative; choose processing parameters for your recording and analysis.

from copy import deepcopy
import numpy as np
from ibo_biomech import MarkerData, TrialData

time = np.arange(300) / 100.0
marker = MarkerData(
    name="R_Ankle",
    x=1000.0 + 50.0 * np.sin(2 * np.pi * time),
    y=np.zeros_like(time),
    z=np.full_like(time, 100.0),
    unit="mm",
    sampling_rate=100.0,
    time=time,
)
trial = TrialData(name="demo", markers={marker.name: marker})
processed = deepcopy(trial)
processed.lowpass_filter_markers(cutoff_freq=6.0, order=4)
processed.crop("markers", start_idx=50, end_idx=200)

frame = processed.as_df(processed.markers)
frame.insert(0, "time", processed.markers["R_Ankle"].time)
print(processed.name, frame.shape)  # demo (150, 4)
# processed.markers["R_Ankle"].plot()  # requires matplotlib

Filtering, cropping, rotation, and unit conversion modify containers in place. Use deepcopy() to keep a raw trial. Marker arithmetic returns a new marker; EMGData.process_emg() returns an envelope, while also cleaning NaNs in raw data.

Load a recording

Replace the paths with your files. HDF5 support is specific to the institute's schema, not arbitrary .h5 files.

from ibo_biomech import C3DHandler, H5Handler

trial = C3DHandler("walking.c3d").load_data()
# Alternatively:
# trial = H5Handler("walking.h5").load_data()

print(trial.name)
print(trial.get_marker_names())
print(trial.get_force_names())
print(trial.get_analog_names())
print(trial.marker_rate, trial.analog_rate, trial.force_rate)

# Use labels present in your recording:
# marker = trial.markers["R_Ankle"]
# plate = trial.forces["forceplate_0"]
# analog = trial.get_analog_by_channel(3)

A C3D handler's returned trial is a separate copy. Processing that trial does not update the handler's raw C3D structure. For processed TRC/MOT output, use the export tutorial.

Current processing API

There is no single trial-wide lowpass_filter(), rotate_data(), or convert_to_meters() method. To crop the same interval across sampling rates, select indices using each channel's time vector; see the processing tutorial.

File conversion and results

FileConverter.c3d_to_h5(c3d_path, h5_path, **metadata) writes an HDF5 file in acquisition coordinates and units. It currently requires analog channels.

OpenSim converters are h5_to_trc, h5_to_mot, h5_to_opensim, c3d_to_trc, c3d_to_mot, and c3d_to_opensim. Their defaults are axis="x", angle=-90, and convert_to_meters=True. Those defaults describe one lab convention. The combined converters accept (source_path, mot_path, trc_path).

The current HDF5 geometry mismatch also affects the convenience force converters. The export tutorial shows a working route from a processed C3D trial to TRC/MOT using write_trc() and write_mot().

from ibo_biomech import IKResults, IDResults

ik = IKResults(filepath="walking_IK.mot")
ik.to_rad()
print(ik["hip_flexion_r"].data)
ik.write("walking_IK_radians.mot")

id_results = IDResults(filepath="walking_ID.sto")
print(id_results.columns)

Check per-column units when reading results: the current shared reader assigns angle units to all columns, including translations and ID forces/moments. See results and subjects for explicit units, DataFrames, normalization, and multi-trial organization.

Tutorials and reference

Start with the documentation overview:

• Load and organize trials: C3D, HDF5, metadata, ownership, and saving constraints.
• Process markers, forces, and EMG: synthetic examples, virtual markers, and time-aware cropping.
• Export OpenSim inputs: processed TRC/MOT output and converter signatures.
• Work with results and subjects: Data, IK/ID, normalization, and tables.
• Run OpenSim tools: scaling, IK, external loads, and ID.
• Detect gait events: an executable example and the detector's assumptions.

API reference pages are in docs/api/. To build the documentation locally:

python -m pip install -e ".[docs]"
python -m sphinx -b html docs docs/_build/html

Open docs/_build/html/index.html. To run the existing tests:

python -m pip install pytest
python -m pytest tests