基于视觉表情分析的交互式表情机器人系统研究

Research on Interactive Expression Robot System Based on Visual Expression Analysis

针对表情机器人双向情感交互的需求,从系统集成的角度提出了一种综合表情分析、识别及交互动作映射的表情机器人系统建模方案.首先,以开源Fritz仿人头部机器人为基础,通过扩展语音、视觉等功能模块,构建了表情机器人多通道情感表达映射模型.其次,为实现表情机器人与操作者的自然交互,提出以视觉分析处理作为情感分析及交互操作的核心,采用类Haar特征级联分类算法进行人脸自动定位检测,利用对光照和人脸姿态变化不敏感的Gabor特征作为情感表情特征并实施空域降采样.针对传统特征脸检测方法样本间分布不合理问题,提出结合类内PCA分析、能量维数及特征重构来筛选特征向量,并张量成特征空间完成表情的分类识别.最后,在JAFFE库及实时视频图片上进行的测试结果证实了该系统建模方案的可行性和有效性.

To meet the need of expression robot＇s mutual emotion interaction,a comprehensive expression robot system scheme was modeled by combining expression analysis, recognition and interactive action mapping. Firstly, based on the open source of Fritz humanoid head robot, multi-channel emotional expression mapping model was proposed by extending the functional module of voice and vision. Secondly, to achieve the natural interaction between the expression robot and the operator, visual analysis is regard as the core of emotion analysis and interactive operation. Haar-like featured cascade classified algorithm is utilized for automatic face location. Af- ter the face detection, Gabor feature is filtered and down-sampled as the emotional expression feature, which is not sensitive to the change of light and human facial posture. To overcome unreasonable distribution among samples in traditional Eigenface, energy di- mension and feature reconstruction are served as rules of selecting eigenvectors in with-in class PCA. The selected eigenvectors are further to built a Eigen-space, which is to accomplish the final classification and recognition of expression. Finally, the experimental results on JAFFE database and real-time video image demonstrate the feasibility and effectiveness of proposed system modeling scheme.