設計與實作適用於社交互動分析的穿戴式平台

Abstract

隨著通信技術的發展，穿戴式感測網絡已經普遍在許多應用中也逐漸滲透到我們的日常生活。從分析穿戴式裝置的感測資訊可以幫助研究人員辨識個人社交行為。藉由個人社交行為互助合作來推斷社群互動情形，例如使用者對話情形這是一種社群互動。在這篇文章中，我們提出了穿戴式平台來截取社交相關的高階資訊，例如 : 講話片段、情緒、社交相關的手部動作與頭部動作，並且整合多種感測元件來進行社群互動分析，例如MEMS加速計、指南針、羅盤、陀螺儀和麥克風。

我們設計了一個有效率且容易使用的穿戴式平台，此系統採用短距離通信的藍牙無線技術進行資料傳輸。穿戴式平台使用時間同步協議來保證穿戴感測網路中的資料時間序列性並沒有任何傳輸延遲。透過語言與非語言交流來實現通用型且高階的資訊萃取的方法。藉由知識分享個人的高階的資訊來分析對話群組關係，這些高階的資訊有助於社交互動分析人員來辨識社交分群的問題。我們的整合系統實作再穿戴式設備上，並且從我們現實生活中有2至10個用戶收集感測資訊來驗證出高階的資訊。

With the evolution of communication technology, wearable sensing networks have been widely adopted in many applications. Wearable device has gradually penetrated into our daily lives. However, it observes and analysis sensor information that can help researcher to recognize social behavior. Furthermore, they collect social behavior to infer sociality in collaborate networks. For example, whom a user has analysis talked with, we call that is a kind of social interaction. This work proposed a wearable platform that extract high-level info. These info are social behavior such as speaker turn, emotion, social-related hand motion and social-related head motion. It integrates multiple deep sensor inferences that span key aspects of everyday life such as MEMS accelerometer, compass, gyroscope and microphone.

To implement the easy-to-use wearable platform in this work. We adopt short-range communication of Bluetooth wireless technology that is suitable on our environments. The key of the design the system platform is guaranteed the sensing time-series data without packet delays. Our system adopt the Timing-Sync Protocol for Sensor Networks (TPSN) that consider the message delay (propagation delay) differently between devices. We adopt exist acoustic and nonverbal sensing approach to realize our general-purpose high-level info. These info help social interaction researcher to recognize grouping problem. After that our system collaborative high-level info to infer conversation group and group context by knowledge sharing. Our system experiment collect sensor data from wearable platform in real life with 2 to 10 users to validate out the high-level info.