UCLA

The need for accurate measures to evaluate and study human expression has grown exponentially in recent years, especially with the proliferation of video and image content across the Internet landscape. However, the study of nonverbal communication still often relies on the creation of data by hand, with humans manually labeling video footage or images. This thesis explores automation of the process through computer vision and machine learning, allowing for better speed and precision. The developed automated classification pipeline is run on video footage of the first and third presidential debates between Donald Trump and Hillary Clinton to gauge its accuracy. Results show the automated pipeline is viable as an easily upscaled replacement for human work, able to both accurately reproduce the results of human labeling of the footage and allow for insight into the various nonverbal idiosyncrasies of the speakers.