多源异构信道信息驱动下多粒度动作感知及多维度情感计算研究

探索真实环境下动作无源感知及情感实时计算的模型与方法，对由此引出的理论问题及关键技术开展深入研究。首先，针对当前有源感知研究中设备依赖、空间受限、粒度粗糙等问题，开展多源异构信道数据驱动下多粒度动作无源感知研究。在理论层面，建立信号-动作映射模型，统一刻画受限物理空间下动作由于衰减、多径对信号产生的影响；在方法层面，针对多源信道数据异构性问题，研究融合特征驱动下多粒度动作感知方法。其次，针对当前情感计算研究中场景限制、维度单一、计算复杂等问题，开展多粒度动作数据驱动下多维度情感实时计算研究。在理论层面，建立动作-情感迁移模型，刻画多维度情感状态空间下，多粒度行为与情感之间的迁移关系；在方法层面，针对多粒度动作数据语义鸿沟问题，研究融合语义驱动下多维度情感计算方法。最后，研制动作-情感透明感知空间示范系统，构建行为-情感大数据库，为公共安全、智能家居、智慧医疗等领域建立研究平台与数据基础。

This project intends to study the sensorless motion sensing and realtime emotion computing issues, including their theoretic problems and key technologies. Firstly, in order to address critical limitations like device-dependent, space/granularity constraints of current research on motion sensing, it leverages heterogeneous channel state info (CSI) extracted from different sources to tackle the sensorless multi-granularity motion sensing issue. The theoretic foundation is to build a signal-motion mapping model to characterize the relationship between motions and signal distortions due to multi-path and fading in terms of CSI. Also, to deal with the heterogeneous data issue, we will propose a novel feature extraction method to obtain unified features, based on which we can distinguish multi-granularity motions. Secondly, in order to address major demerits like high complexity, scenario/dimensional constraints of current research on emotion computing, we explore multi-granularity motions to tackle the realtime emotion computing issue. The theoretic foundation is to build a motion-emotion migrating model to characterize the relationship between motions and emotions in the multi-dimension emotion space. Also, to deal with the heterogeneous semantics issue, we will design a novel fusion method to extract fused semantics of multi-granularity motions, based on which we can compute the multi-dimension emotion in realtime. Lastly, we build a prototype to evaluate our sensorless motion sensing and realtime emotion computing system. Moreover, the prototype will be used to take personalized activity and emotion logs for the tested subjects, given that the test subjects’ mental health can be evaluated accurately. By accumulating such logs from different subjects continuously, we aim to build a large motion-emotion database. In that way, we are able to provide not only theoretical foundations for emerging areas such as the intelligent medical services and the mental health, but also offer the real-world data supports to facilitate their research.