392256 Project: Building a Head Gesture Video Dataset (Pj) (WiSe 2026/2027)

Background and Motivation Progress in automated head gesture recognition is fundamentally limited by the absence of large, diverse, and ecologically valid benchmark datasets — existing corpora are predominantly lab-recorded, culturally homogeneous, and restricted to a narrow range of gesture types. This project addresses that gap by constructing an in-the-wild head gesture dataset scraped from publicly available internet video sources, covering six gesture categories (nod, shake, turn, tilt, up-down, waggle) with balanced representation across demographic groups. A video scraping and filtering pipeline will be designed and implemented, followed by automated quality filtering and automatic extraction and clipping of candidate gesture segments. The project provides hands-on experience with data collection methodology and video processing pipelines. Skills and Requirements * Background in computer vision and basic data engineering. * Familiarity with Python; experience with web scraping.