倒車輔助系統之影像分析與合成

Abstract

本論文中實作了一個即時的提供視訊畫面之倒車輔助系統，不同於其他相關的研究中使用一些特殊的裝置或設備擷取汽車額外資訊，我們沒有使用到任何其它的特定感應器或設備，而只使用一台裝在車後的攝影機拍攝車後影像，再利用平面投影轉換將影像校正成一個俯視畫面，使駕駛者可以直接由畫面中了解汽車與停車格間的關係。並配合特徵點的擷取與追蹤，將過去影像和目前影像進行合成接圖，以提供駕駛者一個完整且清楚的環境俯視圖，讓駕駛者在倒車停車時可以更為快速且容易。

In this thesis, a real-time parking assistance system, which provides a top-view video for a driver to refer to, is investigated. Unlike other parking assistance systems that use special equipments or multiple sensors, the system only uses a rear-view camera to capture scene images behind a backing car. The rear-view images are then converted to top-view ones of the vehicle and its surrounding area using homography transformations. Thus, the driver can easily observe relative positions between the vehicle and the parking area. Moreover, a sequence of top-view images can be further synthesized to generate a single, size-extended one by feature point tracking and image registration. Hence, the system is able to display surrounding area outside the current camera view. By using the proposed system, the driver can have a more clear sense of distance and direction for easier and faster parking in a garage or on the street.