Deep Learning for Gait Analysis
Author Information
Author(s): Kjartan Halvorsen, Peng Wei, Fredrik Olsson, Anna Cristina Åberg
Primary Institution: Dalarna University
Hypothesis
Keypoints on the heel can be accurately determined from video recordings using a combination of OpenPose and custom convolutional neural networks.
Conclusion
The study demonstrates that deep learning methods can accurately identify heel keypoints from video recordings, which may enhance clinical gait analysis.
Supporting Evidence
- The median error for the best-performing model was 0.55 cm.
- All tested models had a median error of less than 0.9 cm.
- The method shows potential for clinical use in gait analysis.
Takeaway
This study shows that we can use videos to find important points on our feet when we walk, which can help doctors understand how we move.
Methodology
The study used convolutional neural networks trained on video data to identify heel keypoints from gait recordings.
Limitations
The accuracy of the method may decrease with images that differ significantly from the training set.
Participant Demographics
Participants were aged 40-90, with 45% being women.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website