Table Tennis Shot Classification

Main goal:

The main goal of this project is to classify and distinguish between technically correct table tennis shot to one that is not.

Application:

A real time mobile application used to improve player skill during practice.

Architecture and Description

We processes the raw table-tennis footage by first splitting the video into individual shots, then using MediaPipe to extract a set of 3D body landmarks for every frame in each shot. These time‐ordered landmark sequences are finally fed into a supervised deep learning model we trained.

demo.1.mp4

the raw images are in the following format:

Media1.mp4

Thanks to Arik Shapira, the manager of TT haifa team for letting me film 🏓📸.