以慣性量測元件為基礎之上下樓層人體行為辨識系統

Abstract

近年來，以無線射頻技術為基礎的室內定位系統在精準度上有不少的提升。然而，飄移問題是造成定位誤差的一項重要因素。在多樓層定位中的飄移問題將造成不可接受的重大誤差。因此，在本篇論文中，我們提出了一個以慣性量測元件為基礎之上下樓層人體行為辨識系統。我們提出了數個創新的演算法，藉由比對目前特徵與訓練特徵來辨識人體行為。利用我們的系統，多層樓定位系統可以在沒有改變樓層的人體行為發生時，將定位結果鎖定在相同的樓層，藉此輔助定位結果的修正。如此一來，定位系統中樓層間的飄移問題將得以解決。

The accuracy of RF-based indoor localization has been improved in recent years. However, drifting problem has been an important factor of causing localization inaccuracy. In multi-story indoor localization, drifting problem between floors is not acceptable. In this thesis, we propose a system to detect human behavior of changing floors, including elevator going up/down floors and going upstairs/downstairs using Inertial Measurement Unit (IMU) sensor. We proposed novel algorithms to recognize human behaviors by matching the current pattern with training patterns. By using our scheme, multi-story indoor localization systems are able to fix the positioning results in the same floor when no floor changing behaviors are detected. Therefore we are able to solve the drifting problem of changing floors.