AdaBoost-EHMM算法及其在行为识别中的应用

AdaBoost-EHMM algorithm and application in action recognition

隐马尔可夫模型(Hidden Markov Model,HMM)是一种有效的时序信号建模方法,已广泛用于语音识别、文字识别等领域,近年来也被用于人的行为识别。人的行为序列是一种特殊的时序信号,每类行为往往包含若干帧关键姿势。利用行为序列的这个特点,提出了AdaBoost-EHMM(AdaBoost-Exemplar-based HMM)算法,并将该算法应用于行为识别中。利用AdaBoost的特征选择方法将行为序列中的典型样本逐个选择出来作为HMM观测概率模型的均值,之后融合多级分类器进行行为识别。实验结果证明AdaBoost-EHMM算法在保证算法收敛的同时提高了识别率。

Hidden Markov Model （HMM） is an effective method of modeling time sequence, and has been widely used in speech recognition, character recognition, and in action recognition recently. Human action sequence is one kind of special time sequences. Each action sequence always includes some key poses. So, AdaBoost-EHMM（AdaBoost-Exemplar-based HMM） algorithm is presented and used in action recognition. AdaBoost method is used to select exemplars from action sequences as the mean values of observation probability model. Fusion of multiple classifiers is adopted to classify action sequence. Effectiveness of the proposed approach is demonstrated with experiments.