UC San Diego

In this thesis, we propose a new approach to measure the quality of a serve in tennis. We formulate this as classifying tennis serves videos into amateur and professional category and introduce a novel two-stage architecture based on LSTM to address this task. In the first stage, we use a state-of-the-art CNN model to extract 3D human pose estimates in a temporal context. An LSTM is then used in the second phase to classify the human pose. We also investigate various 3D human pose estimation algorithms, LSTM architectures, and data collection methods and evaluate their performances for classifying tennis serves.