AutoCode: Automated System for Video Annotation of Human Nonverbal Cues

In this report, we present a system that employs computer vision and machine learning methods to automatically detect and annotate human nonverbal cues in video data. Our system uses face tracking to extract low-level facial motion parameters, and a set of support vector machines to detect and classify higher-level nonverbal behavior cues. For all behaviors detected, time-series annotations are created and exported to a standard format. This annotated data of nonverbal cues can greatly benefit researchers of human behavior, e.g., to develop computational models.