WHT Team 4

IMU Data Harmonization

Struggling to integrate IMU datasets?

This pipeline helps standardize column names, align coordinate systems to the ISB standard, and remove gravity. Each step is adjustable, and the final output is a dataset normalized in g units.

Our Process

🔍

A. File Information Check

📖

B. Structural Harmonization

📐

C. Coordinate Standardization

💾

D. Outcome & Execution

Why Harmonize?

IMU datasets are often fragmented across studies due to differences in naming conventions, coordinate systems, sampling rates, and metadata structures. These inconsistencies prevent direct integration and limit large-scale analysis. Harmonization transforms heterogeneous datasets into a unified framework, enabling consistent interpretation, dataset integration, and scalable modeling. By standardizing structure and coordinates, it becomes possible to combine multiple datasets and support more robust and generalizable biomechanical and machine learning applications.

Run Pipeline in Your Browser

Based on WebAssembly (Pyodide) technology, data processing is performed entirely on your local machine. Your data never leaves your browser. More Info

Phase C: Coordinate Standardization

5. Processing Status

Columns have been synchronized. Ready for coordinate transformation and gravity removal.

            Detected Sensors Metadata:
            
                  Location (Segment)
                  ACC Columns (X, Y, Z)
                  GYR Columns (X, Y, Z)

Initial Static Mean Acc

Initial PCA Variance

6. Coordinate Rotation (ISB Standard)

The system automatically estimates sensor orientation from the signal pattern. It identifies the gravity-related axis using accelerometer magnitude, then uses the remaining axes and gyroscope variation to infer the other directions. A rotation matrix is then constructed to align each sensor to the ISB standard.

Standard: Body Segments (Non-foot)

Target ISB Frame: X = Forward, Y = Up (Gravity), Z = Right

Standard: Foot Segments

Target ISB Frame: X = Up (Gravity), Y = Backward, Z = Right

Phase D: Outcome & Execution

Harmonization Complete

Processing Insight: Note: All output acceleration data has been normalized to g units (1g ≈ 9.81 m/s²).
The comparison bars (top) show that the static gravitational component has been removed, bringing the mean acceleration to ~0g. The waveform (bottom) represents the clean linear acceleration used for activity feature extraction.

🔍 Custom Waveform Inspector

Select a specific body segment and define the start/end index range to generate a detailed time-series waveform of the harmonized linear acceleration.

Select Segment

Start Index

End Index

8. Batch Processing

Apply the confirmed synchronization and standardization settings to multiple files in a folder. All files must follow the same sensor configuration.
Note: Please make sure all files are properly renamed before uploading.

Input Folder (Raw CSVs)

Output Folder (Results)

Enable File Restructuring (Rename & Standardize)

How to configure restructuring:
This tool renames your files to the standard format: p-{pid}_s-{session}_a-{activity}.csv. For each field (PID, Session, Activity), choose where the information should come from:

Filename: Extracts data from the existing filename (see modes below).
Folder Path Part: Uses a specific segment of the folder's directory path.
CSV Column: Reads the value directly from a row in the original data file.
Fixed Value: Applies the same manual entry to all processed files.

Filename Extraction Modes:

1. Whole Stem: Uses the entire filename without the extension.
Example: P01.csv extracts P01.

2. Split & Pick Index: Splits the name by a Delimiter and picks a part by its Index.
Example: P01_S02_Walk.csv with _ and Index 1 extracts S02.

3. Regex Capture Group: Advanced mode using () to define what to extract.
Example: -Session-(\\d+) extracts 02.

Folder Path Part Mode:

Extracts values from the folder hierarchy. Each level of the folder path is assigned an Index (starting from 0).
Example Path: /Data/P01/Session_A/walk.csv
• Index 1 extracts P01.
• Index 2 extracts Session_A.

Participant (PID)

Session

Activity

Global Delimiter (if splitting)

Zero-Padding Width (e.g. 2 for '01')

IMU Data Harmonization

Our Process

A. File Information Check

B. Structural Harmonization

C. Coordinate Standardization

D. Outcome & Execution

Why Harmonize?

Run Pipeline in Your Browser

1. Upload Dataset

2. File Information

3. Processing Configuration

4. Column Synchronization

5. Processing Status

Detected Sensors Metadata:

Initial Static Mean Acc

Initial PCA Variance

6. Coordinate Rotation (ISB Standard)

Verification: Before vs After Rotation

A. Static Mean Acc (Before)

B. PCA Variance (Before)

C. Static Mean Acc (After Sync)

D. PCA Variance (After Sync)

7. Gravity Removal

Harmonization Complete

🔍 Custom Waveform Inspector

8. Batch Processing