Revolutionising movement tracking
Machine learning algorithms can revolutionise human movement tracking by introducing improvements in efficiency, simplicity and accuracy.
Current biomechanical data acquisition techniques use marker-based motion capture, which requires complex and expensive multi-camera systems in gait laboratories. This project has investigated the use of OpenPose, a video-based model for pose estimation.
This deep learning model was compared with Vicon Motion Capture, a gait laboratory system, across movements of varying complexity. This use of OpenPose has potential applications in medicine for diagnoses of orthopaedic injuries, and in defence for soft biometrics and passive gait-based identification. The results of our study have demonstrated that, while highly sensitive to camera calibration, OpenPose is simpler, faster and more flexible than its alternative motion capture techniques.
The accuracy of results produced is, however, not yet adequate for real-world purposes, but stay tuned!