This repository contains 369 strides of walking data, a step event detector, and two stride length estimators.
The walking data is collected from six volunteers equipped with leg-and-shoe-mounted EcoIMU. We attach five EcoIMUs to human body and collect motion data, including triaxial accelerations and triaxial angular rates, through BLE at a 125 samples per second data rate.
The step event detector detects HS (heel-strike) and TO (toe-off) events through a neural network constructed with LSTM cells. The input is 6-axis IMU data collected from one left-shoe-mounted EcoIMU.
We implement this detector in StepEvent_LSTM.py.
We propose two stride length estimators, which use Mechanical Model and LSTM, separately, to estimate stride lengths.
- Mechanical Model: use the z-axis of gyroscope data to obtain the angles at joints through integration, and then we combine these angles with leg length and shoe length to calculate stride lengths. We implement this estimator in StrideLength_MechanicalModel.py.
- LSTM: use 6-axis IMU data collected from all of the five leg-and-shoe-mounted EcoIMUs as the inputs to a neural network constructed with LSTM cells. We implement this estimator in StrideLength_LSTM.py.
These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.
- install cuDNN
- install the following packages:
pip install numpy scipy matplotlib keras
pip install --upgrade tensorflow-gpu==1.5.0Clone this repository and run the corresponding programs. The result will be printed in the terminal.
git clone https://github.com/PoHsin-Lin/gait-parameters-analysis-LSTM.git
python3 StepEvent_LSTM.py
python3 StrideLength_MechanicalModel.py
python3 StrideLength_LSTM.py
For the programs that use LSTM, graphs that show the loss and precision/accuracy during traing, validation, and testing can be found in resultSpace folder.
For more details about the methods and the performance, please see the attached thesis.pdf.