Development and validation of an AI-based mobile application for markerless kinematic analysis of Running Prosthetic Feet in paralympic athletes

This thesis presents the development and validation of an Android mobile application based on an artificial intelligence model for shape-based keypoint recognition, specifically designed for the kinematic analysis of the running prosthetic feet (RPFs) used by elite Paralympic athletes. The application was developed by porting an existing desktop software to a mobile platform. The original software was developed by the INAIL Prosthetic Centre in collaboration with the University of Padova and relies on convolutional neural networks (CNNs) based on the CenterNet and Hourglass architectures to detect keypoints on high-resolution videos of J-shaped and C-shaped RPFs. The application provides a clear and accessible user interface that allows orthopedic technicians and engineers to load videos, perform the calibration, visualize the detected keypoints and carry out the kinematic analysis directly on their smartphone.