車載通訊下融合不精準資訊之合作式定位技術

Abstract

車子定位是一件非常具有挑戰的議題，因為城市中存在許多障礙物 (例如 樹、行人、建築物等)阻隔了通訊信號，而這些障礙物有可能隨時都在改變其位 置，往往不可預測，因此我們必須要想出更多方法或技術去應付這個多變的環境以確保定位演算法的穩定性。藉由裝置間互相分享位置資訊，可以增進定位效能，然而由於不同的GPS (Global Positioning System) 裝置和測距工具會根據位置、設計而有不同的精準度，如果我們盲目地使用這些位置資訊並且結合測量距離的資料，合作式定位的效能將會大打折扣。在本論文中，我們針對V2X (Vehicle to everything) 的通訊環境設計一套合作式定位演算法，考慮當定位資訊如:里程表 (Odometer)、GPS和RSS (Received Signal Strength)等不精準時，我們仍可得到穩定的定位品質。為了達到這個目的，這篇論文定義了一個新的參數叫做信心因子 (Confidence factor)去評斷哪些裝置預期能夠提供更好的定位資訊。這個信心因子與周圍裝置的位置精準度和幾何分布有關係。我們把信心因子融入貝式濾波器 (Bayesian filter)中，以調整個別裝置對定位的影響程度，並因此得到更佳的效能。由實驗結果顯示我們的方法能有效地降低不精準的資訊更新，跟傳統方法比較起來，在城市車載模擬環境中可以降低約50.2%的平均錯誤以及提升73.0%的效能穩定性，因此我們提出的合作式定位技術不止能增進定位的準確度也能提升其精密度。

Traditional vehicle localization in urban is a challenge because of the complicated environment (e.g. tree, pedestrian, building, etc.) which blocks communication signals. We should consider more techniques in this changeable situation to ensure the stability of positioning. By sharing positioning information, we could improve the localization performance. However, we know that different GPS (Global Positioning System) devices and raging models will suffer from different precision due to position or design. If this information is blindly used and directly positioned with the ranging data, the performance will be greatly reduced. In this thesis, we focus on designing a cooperative positioning algorithm in the V2X (Vehicle to everything) scenario with the fusion model of the odometer, GPS and RSS (Received Signal Strength) communication device. With this approach, we can get stable positioning performance even if with imprecise measurements. To achieve this goal, this thesis defines the “confidence factor” to judge which information contains a more positive response. This confidence factor is related to the precision of the surrounding device’s position and geometric position. We use this factor to change the influence of the positional update with current observed data and then combine with the Bayesian filter to obtain the estimated position. The results show that our proposed algorithm can effectively reduce imprecise measurement updates. Comparing to the traditional method, our approach can get 50.2% improvement of average error and 73.0% improvement of stability. Thus, we can conclude our proposed algorithm not only improves the accuracy but also increases the precision.