Diffusion models enable zero-shot pose estimation for lower-limb prosthetic users.

Area of Science:

• Biomechanics
• Computer Vision
• Rehabilitation Engineering

Background:

• Quantitative gait analysis is crucial for lower-limb prosthetic users' rehabilitation.
• Markerless motion capture is challenging due to difficulties in detecting prosthetic joints with standard models.

Purpose of the Study:

• To propose a zero-shot method for markerless gait analysis in lower-limb prosthetic users.
• To leverage generative diffusion models to improve prosthetic joint detection.

Main Methods:

• Utilized generative diffusion models (ControlNet) to transform prosthetic limb images into able-bodied representations.
• Applied standard pose estimation models (OpenPose) to synthetic images for keypoint detection.
• Evaluated the method on publicly available videos of transfemoral and transtibial amputees.

Main Results:

• Achieved significant reductions in keypoint coordinate errors: 37% for transtibial and 76% for transfemoral prostheses.
• Enabled identification of gait deviations like reduced knee flexion and altered kinematic timing.
• Demonstrated feasibility of markerless gait analysis for prosthetic users.

Conclusions:

• The zero-shot approach shows potential for personalized rehabilitation using consumer cameras.
• Further validation on diverse prosthetic designs and larger datasets is warranted.
• This method can enhance gait assessment for individuals with lower-limb prosthetics.