以Cross-Ratio為基礎之自動車夜間視覺定位與誤差分析

Abstract

本篇論文當中，我們介紹了以Cross-Ratio為基礎的自動車定位方法。並且提出一個有效率的方式，來估測Cross-Ratio定位方法所造成定位誤差的分析。我們推導了在合理的誤差範圍內，空間定位誤差與一維影像誤差間的線性關係。由於此一性質，我們將其推廣到二維影像誤差，並找出影像參考點之誤差與空間定位誤差的線性轉換關係。此一誤差分析方法可讓使用者容易地選擇影像參考點，以期得到機率上最正確且可信賴的定位結果。除此之外，我們亦提出了在多個影像參考點（兩點以上）有誤差的情況下，一個有效率的空間定位誤差分析方法。這些誤差分析方法能提供量化指標，可協助我們在多個參考點及Cross-Ratio計算的原點中，針對各種選擇去預估及描述可能的誤差分佈狀況。因此本論文所提出之誤差分析方法，將有助於發展穩定且實用的以Cross-Ratio為基礎的定位系統。

In this thesis, we propose an efficient way of analyzing localization error for systems which use cross-ratio for mobile robot localization. We first inspect the linear nature of localization error due to small 1D inaccuracy in image feature measurements. Properties of the localization error due to two dimensional errors of reference image points are then investigated. Such an analysis will be useful for one to choose point features in stereo images to establish the probabilistically most accurate localization system based on cross-ratio. Moreover, we also proposed an efficient way to analyze localization errors considering multiple noisy image points. Such an approach will be very helpful in practice for the development of a reliable localization system.