GhostCoach: Athletic Coaching Through Computer Vision
Course Instructor
Daniel Cliburn
Abstract
GhostCoach is a standalone athletic coaching application designed to help users analyze and improve their sports techniques through computer vision. Users can upload recorded videos of their athletic movements and receive feedback, performance metrics, and actionable recommendations.
The application is built around four core software modules: a user interface module, a data processing module, an analysis module, and an interpretation module. Upon uploading a video, GhostCoach leverages Google's MoveNet computer vision model to overlay a skeletal model onto the user and track key joint positions, body angles, and motion vectors across frames. This raw motion data is then passed to the analysis module, where it is compared against reference data sourced from higher-level athletes to evaluate the user's technique. These results are sent to the interpretation module that then provides the user feedback. Users can also view visualizations of their motion patterns and compare their data against reference recordings directly within the results screen.
GhostCoach runs on Windows, Linux, and macOS, utilizing OpenCV, NumPy, and MoveNet Python libraries. The current version of the application supports pitching a baseball, but further work can be done to create different models of other athletic techniques. The goal of GhostCoach is to help users from any athletic background develop their techniques and receive objective feedback.
GhostCoach: Athletic Coaching Through Computer Vision
GhostCoach is a standalone athletic coaching application designed to help users analyze and improve their sports techniques through computer vision. Users can upload recorded videos of their athletic movements and receive feedback, performance metrics, and actionable recommendations.
The application is built around four core software modules: a user interface module, a data processing module, an analysis module, and an interpretation module. Upon uploading a video, GhostCoach leverages Google's MoveNet computer vision model to overlay a skeletal model onto the user and track key joint positions, body angles, and motion vectors across frames. This raw motion data is then passed to the analysis module, where it is compared against reference data sourced from higher-level athletes to evaluate the user's technique. These results are sent to the interpretation module that then provides the user feedback. Users can also view visualizations of their motion patterns and compare their data against reference recordings directly within the results screen.
GhostCoach runs on Windows, Linux, and macOS, utilizing OpenCV, NumPy, and MoveNet Python libraries. The current version of the application supports pitching a baseball, but further work can be done to create different models of other athletic techniques. The goal of GhostCoach is to help users from any athletic background develop their techniques and receive objective feedback.
