Real-time classification of dance gestures from skeleton animation

摘要: We present a real-time gesture classification system for skeletal wireframe motion. Its key components include an angular representation of the skeleton designed recognition robustness under noisy input, cascaded correlation-based classifier multivariate time-series data, and distance metric based on dynamic time-warping to evaluate difference in motion between acquired oracle matching gesture. While first last tools are generic nature could be applied any gesture-matching scenario, is conceived assumption that input adheres known, canonical time-base: musical beat. On benchmark comprising 28 classes, hundreds instances recorded using XBOX Kinect platform performed by dozens subjects each class, our has average accuracy 96:9%, approximately 4-second recordings. This remarkable given noise from depth sensor.