基于人格特质的生理信号实时情感识别研究

"Human-machine emotional interactive" is a research hotspot of human-machine interactive. It provides information to intelligent decision-making or emotion regulation systems etc., and therefore, it is the prerequisite and foundation for human-machine interactive. Physiological signals, relative to face, voice and gesture and so on, are recorded continuous easily and more objective in indicating the affective state of the users. Therefore, physiological signals are preferable data source of the real-time emotion recognition system. Considering that there might be difference between the affective state of the user and the affective state recognized through the physiological signals, we analyze the difference of emotional reactions of the subjects of five kinds of personalities, and make a weighted correction of the output of the emotion recognition system, and therefore, make the output meeting the feelings of the users more. We intend to analyze several key points of the personality based real-time emotion recognition using physiological signals, mainly include affective button, which continuously records the feedback of the subject on valence and arousal dimensions in data acquisition, and the fuzzy cluster method and integrated cluster coefficient, which find the optimal feature set correlated best to the valence and arousal dimension of the emotion in data analysis stage, and the difference between emotional reactions of the subjects and in the physiological signals of the subjects of different personalities, which is used to correct the output of the emotion recognition system by weight coefficient.

“人-机情感交互”是人机交互的研究热点，而实时识别用户情感是人机情感交互的前提和基础，为智能决策或用户情感调节等应用提供必要信息。相比于表情、声音、姿态等信息，生理信号易于连续获取，并能客观反应用户的情感状态，更适合作为情感实时识别的信号源。考虑到从生理信号中识别的情感与用户内心的情感体验可能有所差异，分析五类人格特质用户情感体验的差异性，并对生理信号情感识别结果进行加权修正，使情感识别结果更加符合用户的情感体验。项目拟研究基于人格特质的生理信号实时情感识别中的几个关键问题，主要包括信号采集阶段用于获取被试连续情感体验反馈的“情感按钮”，数据分析阶段，用于在数据标签不完全准确的条件下准确提取与情感效价和唤起度最优相关特征集的模糊聚类方法和综合聚类系数，及用于实现不同人格特质用户情感识别输出加权的情绪体验差异性分析和情绪生理反应差异性分析等。

可穿戴技术的发展使得生理信号的测量变得简单易行，将智能穿戴技术应用于日常成活中，通过测量生理信号实时识别用户的情绪状态能够帮助用户了解自身的情绪状态，在健康医疗领域有较好的应用前景。.项目对情感生理信号分析中存在的问题进行探讨，提出一些新的方法，建立了日常生活中用户情绪唤醒度实时识别模型，用于监测情绪波动、焦虑等情绪状态。项目研究内容主要包括：1）完成情绪状态下生理信号数据采集，包括建立被试于静止状态下处于高兴、愤怒、恐惧和悲伤四种情绪状态的生理信号数据库，以及生活状态下实时生理信号的连续采集，捕捉到的情绪包括平静、愉悦和愤怒。2）提取与情感维度相关最优特征集，实现用户非依赖的情绪唤醒度评价。针对生理信号的非线性分析问题，提出子空间分割熵验证生理信号的的非线性特征。针对数据标签不精确数据集的模式分析问题，提出基于聚类熵的聚类评价准则和PSO-FCM聚类分析方法。针对用户非依赖情感识别问题，提出标准人体生理基线映射方法。3）分析不同人格特质用户情感反应差异性，提出识别情感唤起度加权调节方法，实现个性化情感识别。4）分析运动对生理信号实时情感识别模型的影响，去除用户活动对生理信号变化的干扰，得到更准确的情绪唤醒度评价模型。