Catch Posture Analyzer  

The application's labelling of my rowing

A machine learning application developed on Google Colab with Keras and TensorFlow for analyzing streamed video of a rower's posture at the catch on an ergometer. The application uses a pipeline of three machine learning models—a pose estimator, a multilayer perceptron, and an image classification CNN—to give a binary classification ("Good" or "Bad" posture) to individual frames of data, which are then passed to the rower in real time.

Check out the Google Colab demo document for a detailed description of the app and a report on my development process.  

A visualization of the pose data used to identify the catch position