應用於室內之行人追蹤系統及其在手持式裝置上實現

Abstract

近年來，為了改進全球衛星定位的缺點，行人導航系統成為一個熱門的研究題目。其中在行人追蹤上，有許多有價值的應用，例如：在博物館或者商場的顧客指引，或者是以全時間無所不在的健康照護為目的的行人步行距離估算等等，也讓行人室內定位逐漸發展起來。 在本研究中，利用智慧型手機實作了應用於室內之行人追蹤系統，當特定室內區域發生走失事件時，可以藉由此系統得知此行人之位置，達到即時找尋的目的。系統整合應用了手機內部的三軸加速度計及電子羅盤去量測行人運動物理量，推得每一個時間的所在位置，並且結合雲端伺服器與智慧型手機互動，作為追蹤查詢功能之實現。其中三軸加速度計使用在行人移動距離估測上，包含了三個階段，第一階段為步伐偵測，藉由檢查加速度資訊並抓取步行樣本，來偵測是否有移動的行為，第二階段為步伐辨識，利用隱性馬可夫模型來辨識出第一階段所抓取的步行樣本為走路步伐還是跑步步伐，目的在於提高第三階段的準確性，第三階段為步距估測，利用倒傳遞類神經網路去推估出步行樣本的距離，藉此獲得行人移動距離；電子羅盤則是用來獲得行人的移動方向。在行人移動距離估測上，準確性可以達到97％以上。

In recent years, to improve the defects of Global Positioning System, pedestrian navigation system has become a popular research topic. There are many valuable applications on the mission of pedestrian tracking for guest guidance in a museum or a shopping mall, or walking distance estimation for the purpose of pervasive healthcare, and make pedestrian positioning system in indoor environments gradually growing. In this thesis, a pedestrian tracking system in an indoor area is designed for smartphones. When someone is missing in a specific indoor area, we can reach the goal of real-time searching and get the position of the missing pedestrian by this system. The proposed system uses accelerometer and electric compass to measure physical changes of human motion, estimate every-time positions. The system also uses cloud servers for data communicate with smart phones. There are three phases for walking distance measuring with accelerometer. The first phase is step detection. It detects a step occurrence of pedestrian behavior through checking acceleration information and grabbing the step patterns. The second phase is step recognition. It recognizes the patterns from the first phase are walk or run step with Hidden Markov models to raise the accuracy ratio of next phase. The third phase is stride length estimation. It estimates the stride length with backpropagation neural network to get pedestrian distances. The electric compass is used to get the moving vector. The experimental results show the accuracy ratio of 97% for walking distance calculation.