Wearable Spine Tracker vs. Video-Based Pose Estimation for Human Activity Recognition

This repo contains the code for my publication available on MDPI.

Repository Setup

1. Create a Virtual Environment

To set up the repository, first create a virtual environment: Using uv is recommended.

uv venv --python python3.11
source .venv/bin/activate
uv pip install -r requirements.txt

2. Install Local Dependencies

To parse the sensor the package flexlib is needed. The wheels for Linux, MacOS and Windows are located in the wheels folder. Install the right version for you os with:

uv pip install wheels/flexlib-1.0.0-cp311-cp311-manylinux_2_34_x86_64.whl

or

uv pip install wheels/flexlib-1.0.0-cp311-cp311-win_amd64.whl

Running the Code

1. Running Python Scripts

To run the Python scripts, activate the virtual environment and execute the desired script:

python ./driver.py -w 1 -s full -m pose -c best

Checkout the script options with the help flag

python ./driver.py --help

2. Running Jobs from jobs.txt

The jobs.txt file contains predefined commands for running experiments. Each line represents a job with specific parameters. You can execute these jobs using parallel processing:

Example of jobs.txt:

python ./driver.py -w 1 -s combined -m flextail -c quick --inclusion_rate 0.5
python ./driver.py -w 1 -s combined -m pose -c quick --inclusion_rate 0.5
python ./driver.py -w 1 -s full -m flextail -c quick
python ./driver.py -w 1 -s full -m pose -c quick

Running Jobs in Parallel

To run all jobs in parallel:

parallel < jobs.txt

This will execute all commands in jobs.txt simultaneously, leveraging multiple CPU cores for faster processing.

Troubleshooting

• Ensure the virtual environment is activated before running any commands.
• If dependencies fail to install, verify the requirements.txt file and manually install missing packages.

Additional Notes

• The repository is structured to support activity analysis and model evaluation.
• Data files are located in the data directory, organized by participant and activity.
• Results are stored in the results_full_* directories for different modalities.