一個在智慧環境下基於Trie資料結構的人類行為建模與辨識

Abstract

在智慧環境下，人類行為建模與辨識是一個非常重要的任務，它可以應用在許多問題，例如遠程健康監測與干涉。為了觀察智慧環境下的使用者行為，一些建模與辨識的技術已被提出。然而，相關技術都忽略了人類在動態環境下執行同一個行為時的易變性。因此，本論文題出一個基於字典樹(Trie)資料結構的人類行為建模與辨識方法來解決這個問題。此方法分為二階段，建模階段利用歷史資料來建立初始行為模型，調整階段將新進資料做行為辨識，並根據新進的資料來調整行為模型。本研究依據觸發感測器的數量來建立兩個模擬案例並驗證，模擬結果證實本文提出的方法在一個模擬案例可達93%的準確率，另一個案例可達94%的準確率。

In a smart environment, determining how to model and recognize human activity is an important issue. If human activity is modeled and recognized correctly, many services, such as the remote healthcare, etc., can be provided. Several technologies related to modeling and recognizing have been proposed. However, most technologies do not address the following issue: People usually perform the same activity differently in a dynamic environment. To address this issue, we propose a human activity modeling and recognizing method based on the Trie data structure. The proposed method is capable of adjusting the human activity model. This method has two phases: In the first phase, the initial human activity model is constructed by using historical data. In the second phase, new input data are classified and the initial model is adjusted accordingly. Based on the number of triggered sensors, two simulation cases are tested. As for cases 1 and 2, the proposed model has up to 93% and 94%accuracy, respectively.