基于词袋模型的人体动作识别方法

HUMAN ACTION RECOGNITION METHOD BASED ON BAG-OF-WORDS MODEL

传统词袋模型构建的词典不稳定,且忽略词向量先后顺序,在用其进行人体动作识别时,识别效果不稳定,尤其对倒序动作识别效果不佳。针对这些问题,提出一种基于时空联合频率直方图实现动作分类的方法。提取肢体关键角度信息,把关键角度的帧间差值作为时间特征描述子;构建稳定的时间词袋与空间词袋,利用其联合频率直方图表示动作序列,增强动作时间特性;利用支持向量机(SVM)实现动作分类。在一个具有挑战性的数据集-UTKinect数据集上进行实验,结果表明,相比于传统词袋模型与一些已有方法,该方法能够有效提高动作识别的准确率。

The dictionary constructed by the traditional bag-of-words model is unstable and ignores the sequence of word vectors.When it is used for human action recognition,the recognition effect is not stable,and the recognition effect of reverse order action is not good.To solve these problems,a method based on space-temporal joint frequency histogram to realize action classification was proposed.The method extracted the key angle information of the limb,and used the inter-frame D-value of the key angle as the temporal feature descriptor.A stable time bag of words and space bag of words were constructed,and used their joint frequency histogram to represent the action sequence to enhance the action time characteristics.The support vector machine(SVM)was used to implement action classification.Experiments were conducted on challenging UTKinect data set.The results show that compared with traditional bag-of-words models and some existing methods,this method can effectively improve the accuracy of action recognition.