/fidgetyfind

Primary LanguagePython

FidgetyFind

Code for:

Morais, R., Le, V., Morgan, C., Spittle, A., Badawi, N., Valentine, J., Hurrion, E. M., Dawson, P. A., Tran, T., & Venkatesh, S. (2023). Robust and Interpretable General Movement Assessment Using Fidgety Movement Detection. IEEE Journal of Biomedical and Health Informatics, 27(10), 5042–5053.

Skeleton Detection

Setup the OpenPose detector by following the instructions in the fine-tuned OpenPose model for infants. The link is for the parallel branch.

After setting everything up, you can run it on your video:

python -W ignore demo_video.py single_video --video <PATH-TO-YOUR-VIDEO-FILE> --save_root_dir <PATH-TO-SAVE-DIRECTORY>

If you have issues with the script using up all your processors, try:

taskset --cpu-list 0-5 python -W ignore demo_video.py single_video --video <PATH-TO-YOUR-VIDEO-FILE> --save_root_dir <PATH-TO-SAVE-DIRECTORY>

FidgetyFind Environment Setup

Set up the conda environment with:

conda env create -f environment.yml
conda activate fidgetyfind

FidgetyFind Feature Extraction

With the skeleton extracted and the FidgetyFind conda environment setup and on, run:

python ./scripts/fidgetyfind-single-video.py --video_filepath <PATH-TO-YOUR-VIDEO-FILE> \
--skeletons_dir <PATH-TO-DIRECTORY-WITH-DETECTED-SKELETONS> --save_root_dir <PATH-TO-DIRECTORY-TO-SAVE-FEATURES>

The above script will save FidgetyFind features of the hips, hands, and feet to the specified directory.