Continuous Body and Hand Gesture Recognition for Natural Human-Computer Interaction: Extended Abstract

Morency et al. [2007] presented Latent-Dynamic Conditional Random Fields (LDCRF) for We present a new approach to gesture recognition offline sequence labeling and segmentation, but it assumed that tracks body and hands simultaneously and a bounded input sequence, limiting its practical use for online recognizes gestures continuously from an unsegmented gesture recognition. Our work extends LDCRF to an and unbounded input stream. Our system online domain by incorporating our novel multi-layered filtering estimates 3D coordinates of upper body joints technique with a temporal sliding window. We demonstrate and classifies the appearance of hands into a set of our system on the NATOPS aircraft handling signals canonical shapes.