Home
Welcome to the KielMotionAnalysisToolbox (KielMAT). We are a Python based toolbox for processing motion data.
The toolbox is aimed at motion researchers who want to use Python-based open-source software to process their data. We have implemented validated algorithms in modules to process motion data, as shown in the table below:
Overview of modules
| Module | Description | Data |
|---|---|---|
| Gait sequence detection (GSD) | Detects gaits | 3D accelerations from the lower back |
| Initial contact detection (ICD) | Detects initial contact during gait | 3D accelerations from the lower back |
| More to follow... | Additional modules to be added |
The idea is that various motion data can be loaded into our dedicated dataclass which rely on principles from the Motion-BIDS standard.
Installation
The toolbox has been released on pypi and can be installed via pip:
pip install kielmat
Data classes
Data classes: conceptual framework
Motion data is recorded with many different systems and modalities, each with their own proprietary data format. KielMAT deals with this by organizing both data and metadata in a BIDS-like format. The BIDS format suggests that motion recording data from a single tracking system is organized in a single *_tracksys-<label>_motion.tsv file.
[!NOTE] A tracking system is defined as a group of motion channels that share hardware properties (the recording device) and software properties (the recording duration and number of samples).
In KielMAT, data from a single tracking system is therefore loaded into a single pandas.DataFrame. The column headers of this pandas.DataFrame refer to the channels, and the corresponding channels information is likewise available as a pandas.DataFrame.
Similarly, if any events are available for the given recording, these are loaded into a single pandas.DataFrame for each tracking system as well. The events derived from the toolbox can be exported to a BIDS like '*_events.tsv' file.
Data classes: in practice
These concepts are translated into a KielMAT dataclass for each recording: KielMATRecording:
classDiagram
class KielMATRecording {
data: dict[str, pd.DataFrame]
channels: dict[str, pd.DataFrame]
info: None | dict[str, Any] = None
events: None | dict[str, pd.DataFrame] = None
events_info: None | dict[str, Any] = None
add_events(tracking_system, new_events)
add_info(key, value)
export_events(file_path, tracking_system=None, file_name=None, bids_compatible_fname=False)
}
KielMATRecording.data) are organized as a dictionary where the dictionary keys refer to the tracking systems, and the corresponding values the actual (raw) data as a pandas.DataFrame. The description of data channels (KielMATRecording.channels) is availabe as a dictionary with the same keys, and the values contain the channels description.
Example 1: Custom Dataset
You can create a KielMATRecording instance from your own motion data. Suppose you have motion data from your tracking system in a CSV file structured as follows:
timestamp YourTrackedPoint_ACCEL_x YourTrackedPoint_ACCEL_y YourTrackedPoint_ACCEL_z YourTrackedPoint_GYRO_x YourTrackedPoint_GYRO_y YourTrackedPoint_GYRO_z
0 0.00 0.1 0.2 9.8 0.01 0.02 0.03
1 0.01 0.1 0.2 9.8 0.01 0.02 0.03
... ... ... ... ... ... ... ...
You can create a KielMATRecording as follows:
import pandas as pd
from kielmat.utils.kielmat_dataclass import KielMATRecording
# Load your motion data into a pandas DataFrame
motion_data = pd.read_csv("path_to_your_data.csv")
# Calculate the sampling frequency using the timestamp column
time_diff = motion_data["timestamp"].diff().dropna() # Calculate time differences
sampling_frequency = 1 / time_diff.mean() # Sampling frequency in Hz
# Drop the timestamp column, as it is not needed in the motion data
motion_data = motion_data.drop(columns=["timestamp"])
# Define the tracking system and tracked point names
tracking_system = "YourTrackingSystem"
tracked_point = "YourTrackedPoint"
# Create the data dictionary
data = {tracking_system: motion_data}
# Create the channels DataFrame
channels_info = pd.DataFrame({
"name": [f"{tracked_point}_ACCEL_x", f"{tracked_point}_ACCEL_y",
f"{tracked_point}_ACCEL_z", f"{tracked_point}_GYRO_x",
f"{tracked_point}_GYRO_y", f"{tracked_point}_GYRO_z"],
"type": ["ACCEL", "ACCEL", "ACCEL", "GYRO", "GYRO", "GYRO"],
"component": ["x", "y", "z", "x", "y", "z"],
"tracked_point": [tracked_point] * 6,
"units": ["g", "g", "g", "deg/s", "deg/s", "deg/s"],
"sampling_frequency": [sampling_frequency] * 6
})
# Create the channels dictionary
channels = {tracking_system: channels_info}
# Create a KielMATRecording instance
recording = KielMATRecording(data=data, channels=channels)
# Print data and channels information
print(recording.data)
print(recording.channels)
{'YourTrackingSystem': YourTrackedPoint_ACCEL_x YourTrackedPoint_ACCEL_y \
0 0.1 0.2
1 0.1 0.2
... ... ...
YourTrackedPoint_ACCEL_z YourTrackedPoint_GYRO_x YourTrackedPoint_GYRO_y \
0 9.8 0.01 0.02
1 9.8 0.01 0.02
... ... ... ...
YourTrackedPoint_GYRO_z
0 0.03
1 0.03
... ... }
{'YourTrackingSystem': name component type tracked_point units \
0 YourTrackedPoint_ACCEL_x x ACCEL YourTrackedPoint g
1 YourTrackedPoint_ACCEL_y y ACCEL YourTrackedPoint g
2 YourTrackedPoint_ACCEL_z z ACCEL YourTrackedPoint g
3 YourTrackedPoint_GYRO_x x GYRO YourTrackedPoint deg/s
4 YourTrackedPoint_GYRO_y y GYRO YourTrackedPoint deg/s
5 YourTrackedPoint_GYRO_z z GYRO YourTrackedPoint deg/s
sampling_frequency
0 100.0
1 100.0
2 100.0
3 100.0
4 100.0
5 100.0 }
Example 2: Mobilise-D Dataset
You can also load data from the Mobilise-D dataset, one of the datasets available in the toolbox. To do this, use the load_recording() function in the kielmat.datasets.mobilised.
import numpy as np
from pathlib import Path
from kielmat.datasets import mobilised
# Load data from the Mobilise-D dataset
recording = mobilised.load_recording()
# The keys of the recording
recording.__dict__.keys()
dict_keys(['data', 'channels', 'info', 'events', 'events_info'])
# Print the data information
print(recording.data)
{'SU': LowerBack_ACCEL_x LowerBack_ACCEL_y LowerBack_ACCEL_z \
0 0.933334 0.084820 -0.302665
1 0.932675 0.084844 -0.300591
2 0.932350 0.082886 -0.310576
3 0.929716 0.081786 -0.303551
4 0.932825 0.077879 -0.308859
... ... ... ...
693471 -0.192553 -0.016052 -0.984290
693472 -0.189575 -0.016449 -0.988130
693473 -0.191176 -0.017954 -0.983820
693474 -0.189691 -0.014539 -0.986376
693475 -0.192993 -0.015306 -0.989452
LowerBack_GYRO_x LowerBack_GYRO_y LowerBack_GYRO_z \
0 5.600066 1.120697 0.489152
1 5.440734 1.401663 0.279477
2 5.196312 1.168802 0.435765
3 5.553083 1.116346 0.383447
4 5.437505 0.892803 -0.150115
... ... ... ...
693471 -0.225874 0.832856 0.704711
693472 -0.393438 0.598116 0.522755
693473 -0.430749 0.417541 0.282336
693474 -0.279277 0.559122 0.418693
693475 -0.563741 0.478618 0.411295
LowerBack_MAGN_x LowerBack_MAGN_y LowerBack_MAGN_z \
0 -93.972011 -25.023998 44.675028
1 -93.958012 -25.016007 44.610055
2 -93.946010 -25.000014 44.520078
3 -93.938007 -24.980018 44.411097
4 -93.935003 -24.957021 44.287113
... ... ... ...
693471 -50.718928 -36.997006 34.111960
693472 -50.649929 -37.003005 34.072972
693473 -50.579936 -37.008003 34.044986
693474 -50.515946 -37.011000 34.031004
693475 -50.460961 -37.010996 34.035025
LowerBack_BARO_n/a
0 990.394600
1 990.395100
2 990.395600
3 990.396199
4 990.396700
... ...
693471 990.204600
693472 990.204900
693473 990.205200
693474 990.205500
693475 990.205800
[693476 rows x 10 columns]}
# Print the channels information
print(recording.channels)
{'SU':
name type component tracked_point units sampling_frequency
0 LowerBack_ACCEL_x Acc x LowerBack g 100.0
1 LowerBack_ACCEL_y Acc y LowerBack g 100.0
2 LowerBack_ACCEL_z Acc z LowerBack g 100.0
3 LowerBack_ANGVEL_x Gyr x LowerBack deg/s 100.0
4 LowerBack_ANGVEL_y Gyr y LowerBack deg/s 100.0
5 LowerBack_ANGVEL_z Gyr z LowerBack deg/s 100.0
6 LowerBack_MAGN_x Mag x LowerBack µT 100.0
7 LowerBack_MAGN_y Mag y LowerBack µT 100.0
8 LowerBack_MAGN_z Mag z LowerBack µT 100.0
9 LowerBack_BARO_n/a Bar n/a LowerBack hPa 100.0,
}
[!NOTE] In the examples you find a tutorial (Load data into KielMAT) that explains the basics of the dataclass and how to work with them.
Contributing
We welcome contributions to KielMAT! Please refer to our contributing guide for more details.
Paper
The paper has been recently published in JOSS. You can find the paper here.
Authors
Masoud Abedinifar, Julius Welzel, Walter Maetzler, Clint Hansen & Robbin Romijnders
