HMM-based Detection of Head Nods to Evaluate Conversational Engagement from Head Motion Data
HMM-based Detection of Head Nods to Evaluate Conversational Engagement from Head Motion Data
复制标题
基于 HMM 的点头检测，根据头部运动数据评估对话参与度
DOI：
10.23919/eusipco54536.2021.9615999
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
P. Cosman
中科院分区：
文献类型：
--
作者：
Saygin Artiran;L. Chukoskie;Ara Jung;I. Miller;P. Cosman
点头和摇头等头部姿势在谈话中起着重要作用，表明积极倾听和对谈话感兴趣。我们的目标是创建一个工具来评估这些物理会话参与线索的背景下，模拟工作面试。我们提出了一个隐马尔可夫模型为基础的架构，定位和分类的头部摆动和摇头头部运动数据在网上的时尚。基于检测到的头部姿态的数量、速度和持续时间，我们使用线性回归模型来评估会话参与度。对于面试部分，模型分数和人类评分员的分数之间达成了高度一致。我们认为这个系统是一条通往增强现实和基于虚拟现实的培训的道路，可以通过竞争激烈的招聘场景扩大对职业的参与。
Head gestures such as head nodding and shaking play a prominent role in conversation, indicating active listening and interest in a conversation. We aim to create a tool to assess these physical conversational engagement cues in the context of a mock job interview. We propose a hidden Markov model-based architecture to locate and classify head nods and shakes from head motion data in an online fashion. Based on the number, velocity, and duration of the detected head gestures, we evaluate the conversational engagement level using a linear regression model. For the interview segments, high agreement was reached between model scores and scores from human raters. We consider this system as a path toward augmented reality and virtual reality-based training that can broaden participation in careers with competitive hiring scenarios.