運用多感測器於日常生活行為辨識之研究

Utilizing Heterogeneous Sensors for Everyday Activity Recognition

顏嘉楠(Chia-Nan Yen)

指導教授 ： 許永真

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2007年)

https://doi.org/10.6342/NTU.2007.01583

摘要

人口結構的老化儼然已經成為世界各國共同的趨勢。智慧型照護系統研究的主要目的在於設計發展人性化技術來滿足居家老人生活照顧與安全上的需求，而行為辨識即是成功建置老人照護系統的關鍵技術之一。

　　此研究所設計的行為辨識系統採取機率推論，利用多種異質感測器(heterogeneous sensors)來偵測使用者與週遭環境之種種互動，用以辨識其日常生活行為。在此研究中，我們利用感測器捕捉了聲音、位置、使用者與週遭物品之互動等三種資訊。此外，本篇論文並介紹我們所提出的「階層式行為辨識架構」(hierarchical activity recognition framework)。 在這個架構中，我們先針對各感測器偵測出較具意義的事件(semantic sensor events),再進一步結合偵測出的事件以推論出使用者正在從事的行為.

　　我們藉由一個居家環境收集使用者從事日常生活的資料。實驗結果顯示，結合多種異質感測器可有效地將行為辨識的準確率從59.52% 提升到 66.03%。此外，透過「階層式行為辨識架構」，準確率可再進一步地提升至 77.56%。

關鍵字

行為辨識 ； 異質感測器 ； 多感測器 ； 機率推論 ； 隱藏式馬可夫模型 ； 無線射頻識別技術應用

並列摘要

Human life expectancy has increased significantly during the last century. Issues associated with the aging population, such as high medical cost and lack of companionship, have become critical in nearly every country around the world. Intelligent health care systems, which explore advances in information integration, sensory technology, and artificial intelligent to create solutions for "successful aging in place", have attracted much attention. Among all the important technology in building an intelligent health care system, activity recognition is one of the key technologies.

In the field of activity recognition, one of the fundamental questions is how to fuse information from multiple sensors. We confront this question by introducing a hierarchical activity recognition framework, which detects meaningful events from heterogeneous sensors in the first layer and utilizes these detected events for activity recognition in the second layer. In particular, we diagnose the state of a user's activity by integrating information from three types of sensor data: voice, location, and object usage. While various approaches have been tried by past researchers in the field of activity recognition, most of these tend to focus on short-term activities and using only single type of sensor. We suggest that heterogeneous sensors and hierarchical activity recognition frameworks are promising, both in terms of robust performance and reusability.

We collected data in a home-like environment. As the result shows, using heterogeneous sensors can improve the accuracy from 59.52% to 66.03%. Moreover, by using the hierarchical activity recognition framework, the accuracy can be further improved to 77.56%.

並列關鍵字

activity recognition ； heterogeneous sensors ； multiple sensors ； probabilistic reasoning ； hidden Markov model ； RFID application

參考文獻

[1] R. Aipperspach, E. Cohen, and J. Canny. Modeling human behavior from simple

sensors in the home. In Proceedings of the 4th International Conference on

Pervasive Computing (Pervasive 2006), 2006.

[2] L. Bao. Physical activity recognition from acceleration data under seminaturalistic

conditions. Master’s thesis, Electrical Engineering and Computer