Recognition and Segmentation of 3-D Human Action Using HMM and Multi-class AdaBoost

摘要: Our goal is to automatically segment and recognize basic human actions, such as stand, walk wave hands, from a sequence of joint positions or pose angles. Such recognition difficult due high dimensionality the data large spatial temporal variations in same action. We decompose dimensional 3-D space into set feature spaces where each corresponds motion single combination related multiple joints. For feature, dynamics action class learned with one HMM. Given sequence, observation probability computed HMM weak classifier for that formed based on those probabilities. The classifiers strong discriminative power are then combined by Multi-Class AdaBoost (AdaBoost.M2) algorithm. A dynamic programming algorithm applied actions simultaneously. Results recognizing 22 number capture sequences well several annotated tracked show effectiveness proposed algorithms.