中文摘要：

情感识别是人机交互领域的重要研究课题之一，随着研究通道的增多，研究成本和工作量也越来越大。本文在7种模式下通过未受训的观察员来检测人机交互过程中人类的基本情感状态（包括：厌恶、恐惧、快乐、悲伤和惊讶）的自然表达，并评估稳定性。用混合效应逻辑回归模型对2个观察员（oo）之间的一致性进行计算和分析，结果显示一致性普遍偏低。除了比较单模态和多模态的整体一致性，还比较了单个情感状态在单个模式下的一致性，而比较结果则用超可加性、可加性、冗余性和抑制性效应进行分类。目前，自动情感检测结果的意义还在研究中。

英文摘要：

Emotion perception is one of the most important research topics in the field of human-computer interaction. With the increase of channels, research costs and the workload are also increasing. In this paper, the human＇s basic emotion states were been detected by untrained people in seven conditions during human-computer interaction, and assess stability. It computed and analyzed the agreement between two observers（oo） with mixed-effects logistic regression models. The result is generally low. In addition to the overall consistency of the unimodal and multimodal condition, it also compared the consistency of individual affec- tive states in a single model, and classified the results with the superadditive, additive, redundancy and inhibitory effect. The significance of the results of automatic emotion detection is still been discussed.