空間限制對房間定位準確性之研究

Abstract

雖然現在利用訊號強度所做的室內定位系統，其準確度約為兩米左右，但是直接將絕對座標轉換成房間資訊，卻常常造成錯誤。而從事室內定位的專家普遍都有一個共識，就是室內定位系統的準確度必須至少達到房間程度。 本研究結合空間限制的模型Lock Maker與機率模型particle filter，來提昇定位系統在提供房間定位的準確度。藉由particle在不同區域間的分佈情形來更新、給予移動時的空間限制；另外也利用長時間對感測器模型的觀察，協助判斷被追蹤的人是否進入房間，來輔助更新空間限制並適時導引particle移動。 實驗結果顯示，我們所提出的方法在提供房間定位上確實比Walking Area、Collision Detection的方法要值得信賴。雖然會造成2秒左右的進房延遲，但我們覺得這延遲仍在容忍的範圍內，且當定位訊號較穩定時，延遲的情形也可以儘可能的降低。

We discover that it is easy to provide incorrect room-level information which is transformed from the absolute position, although the accuracy of the current RSSI location system is about two meters. Indoor localization researchers have a consensus that systems at least need to provide room-level accuracy. Thus, we propose a spatial constraint model, called Lock Maker and combine it into particle filter to provide better room-level location estimation. In Lock Maker, Space Model constructs the map organization by recording the connection between different regions. Then, Spatial Constraint gives limitations to particles while they are moving. At the same time, we update limitations by observing the weight of particles distribute over different areas. Region-level Sensor Model uses a longer duration of location information to help change limitations in Spatial Constraints and guide particles to a more suitable position than current. In our experiments, we compare Lock Maker with Walking Area approach, and Collision Detection Algorithm. The result shows that Lock Maker is more reliable to provide room-level information. Although this approach causes some delay, but we think that the delay is still tolerable to users.