Action recognition based on motion data is the frontier of computer vision and pattern recognition. However, combined with the results of action recognition,human sentiment extraction is relatively scarce. This paper explores and researches on this aspect. The research starts at the exploration of human emotion dataset. Combining computer graphics, artifical intelligence and machine learning,a standard human emotion dataset can be built. It contains four basic emotions:happiness,anger,sadness,fear,and two derived emotions:surprise and disgust. And it gets the relationship between action and emotion from the perspective of action. According to the divided reference of action recognition, the Period is regarded as the emotion extract unit,to ensure every Period’ s emotion list. Combined with the rate,the emotion of the action can be extracted. Experimental results prove that emotion information can be extracted from 3D action data with the approach.%基于运动数据的动作识别技术是当今计算机视觉和模式识别研究领域的热点问题，然而根据动作识别结果进行人体情绪提取的研究却较少。为此，在现有人体行为学研究的基础上，结合计算机图形学、人工智能和机器学习等技术，得出一套关于人体情绪的数据集合，包括开心、愤怒、悲伤、恐惧4种基本情绪和惊喜、厌恶2种衍生情绪。从动作的角度分析这6种情绪得到动作，即情绪关系，按照动作识别的划分基准，将Period作为情绪提取的最小单位，确定单个Period的情绪列表，再结合动作和速率的参数，提取出在执行该动作时人体的情绪。实验结果表明，该方法能够有效地从3D运动数据中提取情绪信息。
展开▼
