利用智慧型手機上之影像功能及慣性元件來設計行人定位系統之研究

Abstract

在現今的導航當中，GPS和Wi-Fi是很頻繁的應用，但是該設備有其限制，在一些收不到衛星訊號的地方無法使用，如:山洞等。而Wi-Fi定位則需要先建立一些定位點，而並非每個地方都有設置。本論文旨在解決這些定位方法所面對的問題，如何單純的利用手機上所具備的影像功能和慣性元件來進行行人定位。 本論文所研究的場景是在學校的湖畔，該場景是一個較大範圍的區域，且非平坦的區域；相較於使用在室內定位的方法，本系統可以有效抑制因為範圍過大所造成的誤差累積，且可以抵抗部分山坡地形所造成的誤差。 本論文的貢獻有三點：第一，本論文提出基於三維重建路徑模型下的分類樹建立可以有效縮短定位的時間；第二，因為在影像定位時，容易因為多相異視角造成匹配誤差；又因為慣性元件也有誤差累積現象，但慣性元件可以減少影像定位的次數。因此本論文提出了一個基於影像和慣性元件之間互相補償的演算法來達到行人定位更精準更快速的目標；第三，本論文完成一個能實現這個目標的系統雛形，並在校園內驗證成功。

GPS(global position system)and Wi-Fi have been applied in navigation in recent days frequently. However, they are limited in some environment where signal is not available, such as cave. For Wi-Fi , it need some AP to help localization which can not be equipped everywhere. This paper used IMU(inertial measurement unit) and vision function on the smart phone to solve the pedestrian navigation problem. This system did the experiment around the lake at school which is a wide but not flat area. Contrast to other indoor navigation system, this one can decrease the error caused by the long path and sloping fields distance calculation. There are three contributions of this thesis, first, through the KD tree method on the 3D reconstruction model, it can decrease the delay of positioning; Second, various viewpoints of vision positioning can cause match error while IMU may have accumulated error after long distance. In the thesis, it propose a method to compensate vision and IMU by each other. Third, the thesis finished a prototype system to achieve these goals and succeeded in positioning on campus