微表情认知下家用老年陪伴机器人情感状态迁移的研究

The Elderly Companion Robot is an important way to achieve cognitive rehabilitation for the Aged. Facial Micro-expression cognition and home environment provide the vital realization means for enhancing the tight coupling of the perception, cognition, emotion and artificial psychological counseling technology. However, the related study in the world is still in its initial stage, and most of the fundamental theories as well as critical techniques are still focused on discrete emotional state analysis and feature recognition rather than multi-dimensional continuous space affective computing theory. Therefore, the existing research results in this area, in particular the fundamental migration mechanisms are far from adequacy to meet the customers' requirements of Human-Machine-Environment mutual penetration and deep fusion. This project will focus on critical theoretical problems and technical issues of affective state migrations of autonomous control and collaborative responses, which are based on fundamental principles of the association mapping between micro-expression cognitive analysis and affective computing, and the main contents include: 1) construction of micro-expressions & affective mapping mechanism , and 2) the emotional state migration positive regulation mechanism and realization method in the continuous multi-dimensional affective space, and 3)cognitive reappraisal and collaborative optimization in smart home environment. The research results in this project may effectively improve the organic combinations between cognitive mapping analysis and affective computing, achieve the positive regulation in migration process and the global optimization, and therefore facilitate a better cognitive service for the customers.

老年陪伴机器人是实现老人认知康复的重要途径，微表情认知和家居环境为提升其感知-认知-情绪的紧密耦合、人工心理辅导技术提供了关键的实现手段。然而，目前国内外学术界对该领域的研究尚处于起步阶段，现有基础理论与关键技术大多是针对解决离散的情感状态分析与特征识别等问题，而对多维连续空间情感计算基础理论、尤其是情感状态的迁移机理研究不足，研究成果不能满足人们对人-机-环境系统互相渗透、深度融合的需求。本项目将基于微表情的认知分析与情感计算之间的关联映射，重点研究并解决情感状态迁移的自主控制与协同响应关键理论与核心技术问题，包括：1）微表情-情感映射机制的构建；2）连续多维情感空间中情感状态迁移的正向调节机制与实现方法；3）智能家居环境的认知重评与协同优化。本项目研究成果有望促进认知映射分析与情感计算的有机结合，实现迁移过程的正向调节及全局最优，从而更好地满足老年用户的认知心理需求。

认知判断和情绪反应是检验老年人心理健康的基础和核心，也是在智能化老年陪护方面亟待解决的关键问题。本项目通过建立基于视觉语义认知的微表情分析模型，提取视觉关注区域的个性化视觉特征，同时采用双流差分自编码网络去除微表情识别中个体身份信息的影响，提取更具有表征能力的微表情特征，将其与个性化视觉特征进行融合，采用特征稀疏编码的方法进行微表情识别，结合生理信号准确理解情感状态。在此基础上，通过Gross认知的情感交互分析，提出一种基于隐马尔科夫的情感状态转移模型，实现类人的情感计算模型，提高老年陪伴机器人的情感控制能力，增强情感行为的连续性和稳定性。同时联合佛山市第一人民医院开展基于老年认知评估的认知交互实验，通过简明精神状态检查量表（MMSE）和蒙特利尔认知评估量表（MoCA）对结果进行评估和验证。本项目进展顺利，量表结果表明本研究具有一定的医用价值，我们的工作在阿尔兹海默症的判断评估方面做出了有益的探索，取得了一定突破。