Preserving Intangible Cultural Heritage Through Augmented Reality and Machine Learning: A Case Study of the One-Bead Netting Stitch

Bead weaving, as a form of Intangible Cultural Heritage (ICH), embodies tacit procedural knowledge transmitted through high-dexterity, fine-motor practice. Existing digital preservation approaches predominantly emphasize artifact documentation, image archives, or stepwise visual overlays, offering limited capacity to capture and formalize the embodied micro-actions that constitute craft expertise. This paper presents a computational framework for modelling bead weaving gestures as structured, classifiable units using egocentric Augmented Reality (AR) sensing and sequential deep learning. Focusing on the One-Bead Netting stitch as a bounded and repeatable workflow, a gesture taxonomy was developed through craft analysis and expert consultation to operationalize core micro-actions for segmentation and supervised modelling. The system was deployed on Microsoft HoloLens 2 to capture lightweight three-dimensional hand skeletal data from two novice participants who completed structured training prior to data collection. Gesture data were recorded from an egocentric perspective, enabling non-invasive tracking of fine-motor weaving actions during natural weaving of the One-Bead Netting stitch. Following a supervised learning pipeline encompassing schema design, skeletal normalization, temporal segmentation, and model training, a hybrid Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) architecture was trained to classify four compulsory bead weaving gestures, achieving 69.22% validation accuracy. A reduced two-class model focusing on the primary stitching gestures achieved 73% validation accuracy. Beyond classification performance, the study demonstrates a reproducible methodology for translating embodied craft practice into computationally interpretable representations. The proposed framework contributes to the development of gesture-level preservation strategies and lays the groundwork for AR-based instructional feedback systems that support the transmission of embodied cultural knowledge within Human-Computer Interaction contexts.